Fix nvdiffrast OOM: set MAX_JOBS=1 before compilation

Dockerfile

@@ -9,13 +9,14 @@ ENV USE_REAL_MODEL=true
 # Ensure NumPy 1.x for CUDA extension compatibility
 RUN pip install --no-cache-dir "numpy<2"
 
+# Set MAX_JOBS=1 BEFORE any CUDA compilation to limit memory usage
+ENV MAX_JOBS=1
+
 # Install nvdiffrast (CUDA rasterizer) - needs GPU, build here
-RUN pip install --no-cache-dir --no-build-isolation git+https://github.com/NVlabs/nvdiffrast.git
+RUN pip install --no-cache-dir git+https://github.com/NVlabs/nvdiffrast.git
 
 # Build CUDA extensions (requires GPU - only part that needs HuggingFace GPU)
-# MAX_JOBS=1 to reduce memory usage during compilation
 WORKDIR /app/FoundationPose
-ENV MAX_JOBS=1
 RUN cd bundlesdf/mycuda && pip install . --no-build-isolation
 
 # Note: mycpp build, weights download, and build deps are already in base image