robustness: vision/depth render off the event loop (deepcopy snapshot), chat auto-reconnect + friendly errors, demo catches connection failures

testbed/bridge.py

@@ -11,6 +11,7 @@ from __future__ import annotations
 
 import asyncio
 import base64
+import copy
 import io
 import math
 from typing import Any
@@ -206,11 +207,18 @@ class RoomBridge(Bridge):
         )
         async def vision() -> VisionOutput:
             tick = self._tick()
-            img, _ = renderer.render(world)
-            buf = io.BytesIO()
-            img.save(buf, format="PNG")
+            # Render off the event loop so frames never stall the bridge under
+            # load; deepcopy keeps the snapshot consistent with this tool call.
+            snapshot = copy.deepcopy(world)
+            img, _ = await asyncio.to_thread(renderer.render, snapshot)
+
+            def _encode() -> str:
+                buf = io.BytesIO()
+                img.save(buf, format="PNG")
+                return base64.b64encode(buf.getvalue()).decode("ascii")
+
             return VisionOutput(
-                png_b64=base64.b64encode(buf.getvalue()).decode("ascii"),
+                png_b64=await asyncio.to_thread(_encode),
                 width=VISION_WIDTH,
                 height=VISION_HEIGHT,
                 tick=tick,
@@ -222,7 +230,8 @@ class RoomBridge(Bridge):
         )
         async def depth(max_depth: float = 10.0) -> DepthOutput:
             tick = self._tick()
-            _, grid = renderer.render(world, max_depth=max_depth)
+            snapshot = copy.deepcopy(world)
+            _, grid = await asyncio.to_thread(renderer.render, snapshot, max_depth)
             return DepthOutput(
                 width=DEPTH_WIDTH,
                 height=DEPTH_HEIGHT,