Upgrade to pydantic v2

[tokebe]

This PR upgrades dependencies to bring reasoner-pydantic to pydantic v2. Changes:

• Python minimum version bumped to 3.9
• Pydantic minimum version bumped to 2
  • parse_obj() is replaced by model_validate()
  • Model.dict() is replaced by model_dump()
  • Validation now occurs by default on model construction, this can be avoided using model_construct() (not recommended)
  • Normalization on model creation now requires a validation context to disable (see docs/NORMALIZE.md)
  • LogLevel is now directly an enum and uses matching case (e.g. LogLevel.INFO)
• Hashing is now deterministic without users having to set PYTHONHASHSEED
• Minor changes to bring models in line with TRAPI 1.5

[tokebe]

Worthy of note: stable_hash() probably needs to be made more efficient-- in my testing using this PR for Retriever implementation it's called more than most other functions in the entire server (though not having much cumulative call time), and may be contributing to Result.model_validate() being slower than expected.

[maximusunc]

Could we move the stable_hash to a new PR? Seems like it's not quite ready for production, plus it seems out of scope of a bump to pydantic v2.

[tokebe]

@maximusunc I've removed stable_hash from this PR.

[uhbrar]

Without the stable hash, this looks good. As discussed, the plan is to rebase so we can resolve conflicts and merge this in.

[tokebe]

Rebased to main, passes all tests on my local

[uhbrar]

Looks like there are a few errors with imports in the tests. If we can get those fixed and run black, this should be good!

[uhbrar]

Looks like typing.Self requires python >= 3.11.