[Docker] Update the CUDA version in the default Docker image to 12.8 (from 12.1)

[peterschmidt85]

Fixes #3163

• Build 0.11rc3 on staging and do a minimum test
• Build 0.11 on production
• Update the image version in sources to 0.11

[r4victor]

@peterschmidt85, IIRC, cuda 12.8 drops support for old architectures which includes V100, so these old GPUs should be handled if dstack still supports them without custom images.

[r4victor]

Old architectures are said to only be deprecated but I remember having some issues with 12.8 and V100. Maybe related to pytorch or smth, so worth testing.

Architecture support for Maxwell, Pascal, and Volta is considered feature-complete and will be frozen in an upcoming release.

https://docs.nvidia.com/cuda/archive/12.8.0/cuda-toolkit-release-notes/index.html#deprecated-or-dropped-features

[peterschmidt85]

@peterschmidt85, IIRC, cuda 12.8 drops support for old architectures which includes V100, so these old GPUs should be handled if dstack still supports them without custom images.

@r4victor Interestingly enough, when we use V100, we would automatically fallback to our previous OS image that comes with CUDA 12.2, and if we run our new Docker image with CUDA 12.8, V100 will work - I've just confirmed that.

I propose to merge this PR as is. cc @jvstme

[peterschmidt85]

Building 0.11 nowo via https://github.com/dstackai/dstack/actions/runs/18349808238

[peterschmidt85]

0.11 for Azure is finally built via https://github.com/dstackai/dstack/actions/runs/18377527747/job/52356839044