Render only when shared state changes

Add shared-data and bundle invalidation signals, and have handles publish them automatically for SAB and graph mutations.

Document the raw core worker protocol in a dedicated VitePress site and simplify the glTF import and picking example.

Amp-Thread-ID: https://ampcode.com/threads/T-01a01ff8-b91f-724f-8952-f07c6b5042fd
Co-authored-by: Heaust Azure <heaust.azure@gmail.com>

docs/guide/core.md

@@ -11,7 +11,7 @@ infrequent worker messages                  hot shared mutations
 │ create/delete rows           │            │ transforms           │
 │ allocate/delete object slot  │            │ cameras/materials    │
 │ compile/switch graph         │            │ lights/app data      │
-│ play/pause/set FPS           │            │ info.skipRender      │
+│ play/pause/set FPS           │            │ signals + row data   │
 └──────────────┬───────────────┘            └──────────┬───────────┘
                └─────────────────┬─────────────────────┘
                                  ▼

docs/guide/getting-started.md

@@ -24,7 +24,7 @@ const scene = new Scene(canvas, { hdr: true, fps: 60 });
 await scene.ready;
 ```
 
-Omit `fps` to render as fast as the browser and GPU allow. `Scene` initializes conventional SOA rows and loads one clustered-forward HDR render graph. The core itself still starts with only its eight-float `info` row.
+Omit `fps` to render as soon as shared data changes. `Scene` initializes conventional SOA rows and loads one clustered-forward HDR render graph. The core itself still starts with only its eight-float `signals` row.
 
 ## 3. Add a triangle
 

docs/guide/importing-and-picking.md

@@ -34,7 +34,7 @@ If row allocations relocated since `Picking` was created, refresh its shared des
 await picking.refresh();
 ```
 
-The playground below imports the repository's full LFS-backed `sponza.glb` in the importer worker, hydrates all 138 primitives, frames them with an arc camera, and sends a real ray to the BVH worker. Click the preview to pick again.
+The playground below imports the repository's full LFS-backed `sponza.glb` in the importer worker, preserves its authored transforms, places an arc camera in its coordinate system, and sends a real ray to the BVH worker. Click the preview to pick again.
 
 <Playground example="importing" />
 

docs/guide/scene-and-sab.md

@@ -29,21 +29,23 @@ particles.row(42).set([1, 0, 0, 0]);
 
 Rows are 16-byte-stride-aligned and arena allocations are 64-byte aligned. Formats are `f32`, `u32`, or `i32`.
 
-## Timing and render skipping
+## Timing and render signals
 
-Core always creates `info` as:
+Core always creates `signals` as:
 
 ```text
-[deltaTime, frameCount, elapsedTime, targetFps, skipRender, 0, 0, 0]
+[deltaTime, frameCount, elapsedTime, targetFps, skipRender, sabDirty, bundleDirty, 0]
 ```
 
 ```ts
-const info = scene.array("info").row(0);
-info[4] = 1; // keep timing, skip GPU work
-info[4] = 0; // resume rendering
+const signals = scene.array("signals").row(0);
+signals[4] = 1; // keep timing, skip GPU work
+signals[4] = 0; // resume and request a frame
 ```
 
-Use messages for rare control changes (`setFps`, graph updates, allocation); use SAB writes for existing hot state.
+The handles layer sets `sabDirty` for writes made through `scene.array(...)` and its node, camera, mesh, material, and light APIs. It sets `bundleDirty` before rebuilding a graph whose recorded pipeline, bindings, geometry, or draw commands changed. Core clears `sabDirty` when it starts a frame; switching to the replacement loadout clears `bundleDirty` and requests a fresh frame.
+
+Use messages for rare control changes (`setFps`, graph updates, allocation); use SAB writes for existing hot state. Code using `@yawn/core` directly must set `signals[5] = 1` after completing its own row writes.
 
 <script setup>
 import Playground from "../.vitepress/Playground.vue";