Headless pictures draw on the GPU, with the viewport's pipelines (CPU rasterizer as last tier) (#1555)

Thanks @jyviko.

docs/book/src/dev/architecture.md

@@ -18,5 +18,12 @@ A few rules follow from the layering:
 - **Everything the user can do is a command** in `crates/engine`. See
   [The command engine](commands.md).
 - **The front end is swappable.** Nothing below `ui-egui` depends on egui. The CLI and the MCP
-  server run the same engine without a window, and `crates/render` has a CPU rasterizer so
-  headless screenshots work anywhere.
+  server run the same engine without a window. Headless pictures (`solidcraft-cli snapshot`, MCP
+  screenshots, `ui.render`, thumbnails) are drawn on the GPU with the viewport's own pipelines
+  (`crates/gpu`, `capture`): the engine describes the scene (`view::scene`) and the host installs
+  the renderer (`engine::capture`). The tiers, in order: a hardware GPU, a software adapter (Mesa
+  lavapipe or llvmpipe, WARP), and last `crates/render`'s CPU rasterizer, which draws no surface
+  analysis, highlights or legends. `SOLIDCRAFT_GPU=auto` walks all three, `hardware` and `software`
+  stay on their tier (an error without one), `cpu` forces the rasterizer; every picture's reply
+  names its `renderer` (`gpu`, `gpu-software`, `cpu`) and a CPU picture carries a warning.
+  `solidcraft-cli gpu` lists the adapters and the tier the policy lands on.

docs/book/src/guide/getting-started.md

@@ -186,7 +186,7 @@ A script is a list of commands:
 
 ![The bracket example, rendered by solidcraft-cli snapshot](../generated/bracket.png)
 
-`snapshot` looks at the model from the iso view. `--view NAME` takes another standard view (`front`, `back`, `top`, `bottom`, `left`, `right`) or the name of a view saved in the design with `view.save`. `--camera JSON` takes any camera, as the MCP `screenshot` tool does (its conventions are in [the MCP page](mcp.md#camera-and-close-up-recipe)). A script that applies a view itself with `view.apply` sets the camera too. The ground grid lies on Z 0, so it cuts through a part that straddles it: `--no-grid` leaves it out of the picture, and `--no-axes` the origin's axes.
+`snapshot` looks at the model from the iso view. `--view NAME` takes another standard view (`front`, `back`, `top`, `bottom`, `left`, `right`) or the name of a view saved in the design with `view.save`. `--camera JSON` takes any camera, as the MCP `screenshot` tool does (its conventions are in [the MCP page](mcp.md#camera-and-close-up-recipe)). A script that applies a view itself with `view.apply` sets the camera too. The ground grid lies on Z 0, so it cuts through a part that straddles it: `--no-grid` leaves it out of the picture, and `--no-axes` the origin's axes. The picture is drawn on the GPU with the viewport's own pipelines, so it shows what the window would (a surface analysis such as `inspect.zebra` included); a machine without a GPU uses a software one (Mesa's lavapipe on Linux, WARP on Windows), and with neither (a Mac without usable Metal, a Linux box without Mesa's Vulkan) the CPU rasterizer draws it, without surface analysis colours, and the reply's `renderer` says `cpu` with a warning. `SOLIDCRAFT_GPU=auto|hardware|software|cpu` chooses (`hardware` and `software` fail rather than fall to the CPU), and `solidcraft-cli gpu` lists the adapters and the tier a snapshot would use.
 
 ```sh
 solidcraft-cli snapshot part.solidcraft --out photo.png --view Photo

docs/control-protocol.md

@@ -136,7 +136,7 @@ labels; `prefs:measure_axis_distances` identifies the checkbox.
 | `ui.key` | `key` (egui key name, `Enter`, `Escape`…), `cmd?`, `shift?` | |
 | `ui.text` | `text` | typed text |
 | `ui.screenshot` | `path?` | PNG of the window (needs a presented frame) |
-| `ui.render` | `path`, `width?`, `height?`, `grid?`, `axes?` | CPU render of the model with the current camera (no window needed), in the window's theme (light or dark; `solidcraft-cli snapshot` keeps the light colours); cut by the section the viewport shows (`inspect.section`, an open Section Analysis dialog's preview, a sketch's Slice), with the cut face capped. The ground grid and the origin's axes are drawn as the window shows them (`ui.set` `showGrid` and `showOrigin`, `hiddenOrigin`); `grid` and `axes` (booleans) say otherwise for this render |
+| `ui.render` | `path`, `width?`, `height?`, `grid?`, `axes?` | Render of the model with the current camera (no window needed), drawn on the app's GPU device with the viewport's pipelines (the reply's `renderer`: `gpu`, or `gpu-software` on a software adapter), in the window's theme (light or dark; `solidcraft-cli snapshot` keeps the light colours); cut by the section the viewport shows (`inspect.section`, an open Section Analysis dialog's preview, a sketch's Slice), with the cut face capped. The ground grid and the origin's axes are drawn as the window shows them (`ui.set` `showGrid` and `showOrigin`, `hiddenOrigin`); `grid` and `axes` (booleans) say otherwise for this render |
 | `ui.confirm` | `accept?`: bool | answers the Delete confirmation (shown when a delete takes other features with it or makes some fail); without `accept`, what it lists |
 | `ui.documents` | `action?`: `new`, `switch`, `close`, `quit`; `index?`; `force?` (close without asking); `answer?`: `save`, `dont_save`, `cancel` (answers "save changes?") | the open designs (tabs): names, unsaved flags, the active one, a pending "save changes?" prompt, whether a quit is asking |
 | `ui.file` | `path`; `insert?`: bool | starts an interactive background Open or Insert; poll `ui.inspect.file_busy` for completion; the window stays responsive |

docs/mcp.md

@@ -18,7 +18,7 @@ document remains the reference implementation today.
 ## Start it
 
 ```sh
-# Headless: an in-process design session, no window (CPU renders for screenshots).
+# Headless: an in-process design session, no window (screenshots are drawn on the GPU).
 solidcraft-cli mcp
 
 # Headless, starting from a design file or a command script.
@@ -149,7 +149,7 @@ They cover the workflow, the command shapes, how to verify with `measure`, and t
 | `measure` | `body?` or `items?` | Volume (mm³), area (mm²), centre of mass, bbox and topology counts per body, with totals. With `items` (one or two selections as `execute` takes them: `face`, `edge`, `vertex`, `sketch_curve`, `profile`), the picked geometry instead: see below. |
 | `body_topology` | `body` | Edges (index, midpoint, length, ends) and faces (index, name, area, centroid, normal, `point`, and for a cylindrical face its `cylinder`: `origin`, `axis`, `radius`). An edge midpoint is the `[x,y,z]` that picks that edge for fillet and chamfer. A face's `point` lies on it and picks it (its centroid may not: that of a face with a hole is in the hole); a face's `name` picks it for the commands that take faces (`face.delete`, `solid.shell`, `solid.draft`, `solid.offset_face`, `solid.split_face`, `solid.replace_face`, and `face` of `sheet.convert`). |
 | `set_parameter` | `name`, `value` (number or expression), `unit?`, `comment?` | Create or change a user parameter and recompute; reports `recomputed` (features rebuilt), `restored` (features whose earlier result for the same inputs was reused, e.g. a value set back) and features that now fail. |
-| `screenshot` | `view?` (iso, top, front, back, bottom, left, right, home, fit, or a named view saved with `view.save`), `camera?` (`{target, yaw, pitch, distance, fov?}`, as `view.apply`), `path?`, `width?`, `height?`, `source?` (window, model), `grid?`, `axes?` | A PNG image content block. Headless: CPU render fitted to the model, or from the named view or camera. Bridged: the app window (default) or a model render with the app's camera; a named view or camera moves the app's viewport there first (`execute` `view.apply` does the same without a screenshot). `grid: false` and `axes: false` leave the ground grid (it lies on Z 0 and cuts through a part that straddles it) and the origin's axes out of a model render; axes hidden with `browser.visibility` stay out whatever `axes` says. Bridged, a screenshot that gives `grid` or `axes` is a model render, which follows the app's Grid and Origin switches for what it does not give; `source: "window"` with either is an error. |
+| `screenshot` | `view?` (iso, top, front, back, bottom, left, right, home, fit, or a named view saved with `view.save`), `camera?` (`{target, yaw, pitch, distance, fov?}`, as `view.apply`), `path?`, `width?`, `height?`, `source?` (window, model), `grid?`, `axes?` | A PNG image content block. Headless: a render with the viewport's pipelines on a headless GPU device (see `solidcraft-cli gpu`; with no adapter, the CPU rasterizer, without analysis colours), fitted to the model, or from the named view or camera; the reply's `renderer` says which (`gpu`, `gpu-software`, `cpu`, with a `warning` for `cpu`). Bridged: the app window (default) or a model render with the app's camera; a named view or camera moves the app's viewport there first (`execute` `view.apply` does the same without a screenshot). `grid: false` and `axes: false` leave the ground grid (it lies on Z 0 and cuts through a part that straddles it) and the origin's axes out of a model render; axes hidden with `browser.visibility` stay out whatever `axes` says. Bridged, a screenshot that gives `grid` or `axes` is a model render, which follows the app's Grid and Origin switches for what it does not give; `source: "window"` with either is an error. |
 | `export` | `path`, `format?` (step, 3mf, 3mf-tagged, stl, stla, obj, urdf), `bodies?`, `mesh_prefix?`, `effort?`, `velocity?` | STEP, 3MF or mesh export; URDF writes the whole assembly with its meshes next to it (`docs/urdf.md`). |
 | `undo`, `redo` | — | Undo / redo one change. |
 | `crash_reports` | `id?`, `limit?` | The crash reports on this computer, newest first; with `id`, the whole report (backtrace, command, recent commands, `design_path`/`replay_path`). See [crash-reports.md](crash-reports.md). |