Use GPUVideoSource for hardware-accelerated video decoding

- CIDVideoSource now uses GPUVideoSource when GPU is available
- Enables CUDA hardware decoding for video sources
- Should significantly improve rendering performance

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

tasks/streaming.py

@@ -134,9 +134,18 @@ class CIDVideoSource:
                 raise ValueError(f"Could not resolve video source '{self.cid}' for actor_id={self.actor_id}")
 
             logger.info(f"CIDVideoSource._ensure_source: resolved to path={path}")
-            # Import from primitives where VideoSource is defined
-            from sexp_effects.primitive_libs.streaming import VideoSource
-            self._source = VideoSource(str(path), self.fps)
+            # Use GPU-accelerated video source if available
+            try:
+                from sexp_effects.primitive_libs.streaming_gpu import GPUVideoSource, GPU_AVAILABLE
+                if GPU_AVAILABLE:
+                    logger.info(f"CIDVideoSource: using GPUVideoSource for {path}")
+                    self._source = GPUVideoSource(str(path), self.fps, prefer_gpu=True)
+                else:
+                    raise ImportError("GPU not available")
+            except (ImportError, Exception) as e:
+                logger.info(f"CIDVideoSource: falling back to CPU VideoSource ({e})")
+                from sexp_effects.primitive_libs.streaming import VideoSource
+                self._source = VideoSource(str(path), self.fps)
 
     def read_at(self, t: float):
         self._ensure_source()