Enforce strict version matches between `jupyter-ai` and `jupyter-ai-magics`

[dlqqq]

Problem

We currently lack an automated way to enforce a strict version match between jupyter-ai-magics and jupyter-ai when installed via pip. Fixes in jupyter-ai sometimes require downstream changes in jupyter-ai-magics, so users may need to update both packages to receive a bug fix. However, this is not done automatically by pip, which is really confusing to the end user.

This has caused several bugs:

• Bug: Installing jupyter-ai==3.0.0a0 alone via pip does not work  #1172
• Bug in v3.0.0a0: Unable to open AI settings #1253
• At least one more. Please feel free to comment below if you find another.

Proposed Solution

• Update bump-version.sh to somehow bump the version pin of jupyter-ai-magics.
• Assert this in the release process somehow to prevent broken releases if this script gets broken.

Additional context

We do enforce this in our Conda Forge releases because that process is manual anyways. However, we should be doing this for PyPI releases too.

[krassowski]

I mentioned it before somewhere else, but jupyter-collaboration bumps the minimum required version in bump_version release scriptr: https://github.com/jupyterlab/jupyter-collaboration/blob/1c23b688ba22c1911d3ee93542e1da6e67f0514a/scripts/bump_version.py#L101-L124

but see discussion in jupyterlab/jupyter-collaboration#343

[krassowski]

Ah yes, #745