Fix missing libnvdla_compiler.so in Jetson 6.x TRT runtime (#2201)

* fix: Add missing libnvdla_compiler.so to Jetson 6.x runtime images

The TensorRT execution provider in ONNX Runtime requires
libnvdla_compiler.so to initialize. Without it, TRT fails to load
and ORT silently falls back to CPU, causing ~3x slower inference.

Affects both JP6.0 and JP6.2 multi-stage Dockerfiles.

* fix: Correct path for libnvdla_compiler.so (tegra subdirectory)

The library lives at /usr/lib/aarch64-linux-gnu/tegra/, not
/usr/lib/aarch64-linux-gnu/ directly.

---------

Co-authored-by: Paweł Pęczek <146137186+PawelPeczek-Roboflow@users.noreply.github.com>

Author: alexnorell

docker/dockerfiles/Dockerfile.onnx.jetson.6.0.0

@@ -279,6 +279,7 @@ COPY --from=builder /usr/local/cuda/targets/aarch64-linux/lib/libnvToolsExt*.so.
 COPY --from=builder /usr/lib/aarch64-linux-gnu/libnvinfer*.so.*.*.* /usr/local/cuda/lib64/
 COPY --from=builder /usr/lib/aarch64-linux-gnu/libnvonnxparser*.so.*.*.* /usr/local/cuda/lib64/
 COPY --from=builder /usr/lib/aarch64-linux-gnu/libnvparsers*.so.*.*.* /usr/local/cuda/lib64/
+COPY --from=builder /usr/lib/aarch64-linux-gnu/tegra/libnvdla_compiler.so* /usr/local/cuda/lib64/
 
 # Create symlinks for cuDNN and TensorRT (saves ~2GB by not duplicating files)
 RUN cd /usr/local/cuda/lib64 && \

docker/dockerfiles/Dockerfile.onnx.jetson.6.2.0

@@ -282,6 +282,7 @@ COPY --from=builder /usr/local/cuda/targets/aarch64-linux/lib/libnvToolsExt*.so.
 COPY --from=builder /usr/lib/aarch64-linux-gnu/libnvinfer*.so.*.*.* /usr/local/cuda/lib64/
 COPY --from=builder /usr/lib/aarch64-linux-gnu/libnvonnxparser*.so.*.*.* /usr/local/cuda/lib64/
 COPY --from=builder /usr/lib/aarch64-linux-gnu/libnvparsers*.so.*.*.* /usr/local/cuda/lib64/
+COPY --from=builder /usr/lib/aarch64-linux-gnu/tegra/libnvdla_compiler.so* /usr/local/cuda/lib64/
 
 # Create symlinks for cuDNN and TensorRT (saves ~2GB by not duplicating files)
 RUN cd /usr/local/cuda/lib64 && \