[Verda] Rename the `datacrunch` backend to `verda`

[peterschmidt85]

Implementation notes:

• Let's keep the datacrunch backend but mark it as deprecated, and introduce verda as another class inherited from the datacrunch backend class.
• In the docs, we only include verda
• In gpuhunt, we simply rename the datacrunch provider to verda

[jvstme]

Let's keep the datacrunch backend but mark it as deprecated, and introduce verda as another class inherited from the datacrunch backend class.

This means there will be two different backends - datacrunch and verda. This solution will require explicit action from all users to migrate from datacrunch to verda, which includes stopping runs, deleting datacrunch instances, and deleting the datacrunch backend.

In dstack Sky, we'll have to either perform this migration on our own (possibly interrupting users' workflows) or also allow having both backends with dstack creds, yet stop creating the datacrunch backend automatically for new projects.

I think I'd prefer datacrunch to be an alias for verda instead, as this would make the migration more seamless. But if that's too much work (it will probably require writing a bunch of validators throughout the codebase to covert datacrunch to verda in all inputs, such as configurations, server configurations, db models, etc), then I don't mind the suggested solution too.

In gpuhunt, we simply rename the datacrunch provider to verda

We can rename it in the codebase, but the offline catalogs should include both datacrunch and verda for compatibility with all dstack versions. Should be as easy as cp verda.csv datacrunch.csv in the CI workflow.

[peterschmidt85]

@jvstme I personally would prefer introducing verda as a separate class and deprecate the datacrunch.
In Sky, existing users will continue use datacrunch until in several months we automatically rename them in their configs for all.
IMO, this is simpler than implementing, testing, and maintaining the new concept of aliases.

Please let me know if you're OK with this.

[jvstme]

Yes, this solution is ok with me.

until in several months we automatically rename them in their configs

Except renaming won't work if somebody creates datacrunch and verda at the same time, in which case the only way to get rid of the datacrunch backend would be to delete it.

[peterschmidt85]

Except renaming won't work if somebody creates datacrunch and verda at the same time

I assume we can try to rename and also disable datacranch at the same time.

[peterschmidt85]

Related PR: dstackai/gpuhunt#205