try out pydantic (or just data classes) for REST API validation

[minrk]

Proposed change

Python dataclasses (and at a higher level, pydantic) allow convenient validation and serialization of JSON request/reply schemas. I think we can try this out incrementally with new APIs and slowly adopt it more as we like it. It should have no user-facing consequences (aside from the dependency, see below).

Alternative options

• Just use dataclasses (stdlib, avoids current issues around the pydantic v2 migration, which will affect users because it would be installed in the single-user environment)
• Keep manual JSON validation (tedious, I started looking into this for user-initiated sharing #4594 after writing yet another JSON model validation method

Who would use this feature?

Mostly JupyterHub maintainers, but especially with pydantic, it could be used to generate accurate docs, as is done with FastAPI.

[minrk]

This is currently done on the sharing endpoints only. I had a pretty good experience, so I think we can iteratively apply these to the other endpoints.