[Feature]: [AWS] Allow to specify AMI identifiers in the backend config

[un-def]

Problem

Currently, it's not possible to use AMIs other than hardcoded in the aws backend, which means that (at least):

• it's not possible to customize instances, e.g., automount some network volumes, configure Docker or use a different Docker version, etc.
• some instance features are not available due to the missing kernel modules
• on NVIDIA instances, users can only use Docker images with the CUDA version compatible with the host CUDA version chosen by dstack (it's 12.1 currently) and only those CUDA libraries that are installed in the dstack-provided AMIs.

Solution

Add a backend option to override the default dstack-provided AMIs.

Workaround

No response

Would you like to help us implement this feature by sending a PR?

Yes