Use a common base image for CUDA libraries

[kenji-miyake]

Checklist

• [X] I've read the contribution guidelines.
• [X] I've searched other issues and no duplicate issues were found.
• [X] I've agreed with the maintainers that I can plan this task.

Description

Currently, we install CUDA using setup-dev-env.sh, but it creates different layers and enforces people to pull a large part of the image every time.

To avoid that, it's better to have a common base image.

Purpose

To minimize the differences between Autoware's Docker Images.

Possible approaches

• Wait for NVIDIA to release a new image and deb packages, and use them. (Recommended)
  • We might have to add the --no-cuda option to setup-dev-env.sh.
• Create a common Autoware base image.

Definition of done

Autoware's Docker Images are based on a common base image.

[kenji-miyake]

For Galactic, it will be resolved by https://github.com/autowarefoundation/autoware/pull/2703.

[xmfcx]

For both arm64 and amd64, we can use nvidia/cuda:12.3.0-base-ubuntu22.04 since now we have upgraded to CUDA 12.3 with:

• https://github.com/autowarefoundation/autoware/issues/3943

I will create a small PR to build and test.

[xmfcx]

Closing this since it will be handled in more detail in:

• https://github.com/autowarefoundation/autoware/issues/3999