Fix intermittent build-cuda-devcontainer failure caused by NVIDIA CUDA apt repo instability

[Copilot]

The build-cuda-devcontainer CI job fails intermittently because the nvidia/cuda base image registers the NVIDIA CUDA apt repo, and when the common-utils devcontainer feature runs apt-get update, it hits that repo during a mirror sync — resulting in a hash mismatch error and exit code 100.

The CUDA toolkit is already bundled in the base image; the apt repo serves no purpose after the Dockerfile RUN steps complete.

Changes

• .devcontainer/Dockerfile: Add a final RUN step that removes NVIDIA CUDA apt source lists and cleans the apt cache, so subsequent apt-get update calls by devcontainer features only hit stable Ubuntu mirrors:

# Remove NVIDIA CUDA apt sources to prevent devcontainer feature installation failures.
# CUDA is already provided by the base image; the apt repo is not needed and causes
# intermittent "mirror sync in progress" errors when devcontainer features run apt-get update.
RUN rm -f /etc/apt/sources.list.d/cuda*.list /etc/apt/sources.list.d/nvidia*.list \
    && apt-get clean