Use PyNvCodec for true zero-copy GPU video decode

Replace decord (CPU-only pip package) with PyNvCodec which provides
direct NVDEC access. Frames decode straight to GPU memory without
any CPU transfer, eliminating the memory bandwidth bottleneck.

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

Dockerfile.gpu

@@ -26,9 +26,9 @@ RUN pip install --no-cache-dir -r requirements.txt
 # Install GPU-specific dependencies (CuPy for CUDA 12.x)
 RUN pip install --no-cache-dir cupy-cuda12x
 
-# Install decord for GPU-accelerated video decoding (keeps frames on GPU)
-# This avoids CPU<->GPU memory transfers during video decode
-RUN pip install --no-cache-dir decord
+# Install PyNvCodec for true GPU-native video decoding (NVDEC)
+# Frames decode directly to GPU memory - zero CPU transfer
+RUN pip install --no-cache-dir PyNvCodec
 
 # Copy application
 COPY . .