feat: per-phase timings of the frame path in the log

One log line per phase every PHASE_LOG_EVERY samples (default 100, 0 = off):
mean and max of wait_frames, align, convert, each JPEG/PNG encoding,
colormap, depth_stats, base64, json and ws_send. Meant for the measurement
on PC2 via `docker logs` (optimize-streams R01): where a capture spends its
~130 ms of CPU is not known yet.

capture() moves its numpy conversion into _frame_from() unchanged, only so
the block can be timed; _depth_colormap() likewise wraps _jet_colormap().

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

src/backend/api/websocket.py

@@ -5,6 +5,7 @@ from fastapi import APIRouter, WebSocket, WebSocketDisconnect
 
 from src.backend.api.routes import _frame_to_dict
 from src.backend.services.image_saver import ImageSaver
+from src.backend.services.phase_timing import timer
 
 try:
     from src.backend.services.detector import Detector
@@ -60,7 +61,8 @@ async def websocket_stream(ws: WebSocket):
                     data = _frame_to_dict(frame)
                     if predicting and _detector is not None:
                         data["detections"] = await asyncio.to_thread(_detector.detect, frame.rgb)
-                    await ws.send_json(data)
+                    with timer.phase("ws_send"):
+                        await ws.send_json(data)
                     if _saver and _saver.is_recording:
                         _saver.save_rec_frame(frame.rgb)
                 await asyncio.sleep(0.033)