Fix JSON serialization with pydantic >=2.12

[michalkrompiec]

Summary

pydantic 2.12 changed serialize_as_any=True to propagate the flag to all nested values (pydantic/pydantic#12348). This broke JSON serialization in two ways, requiring two fixes.

Fix 1: pydantic.py — remove serialize_as_any=True from model_dump_json()

With pydantic ≥2.12, serialize_as_any=True in JSON mode bypasses the Annotated WrapSerializer on _Quantity and tries to serialize pint.Quantity by its runtime type, which has no pydantic schema:

PydanticSerializationError: Unable to serialize unknown type: <class 'pint.Quantity'>

The WrapSerializer(quantity_json_serializer) on _Quantity handles JSON serialization correctly without this flag. serialize_as_any=True is intentionally kept in model_dump() (Python mode), which is unaffected by the pydantic 2.12 change.

Fix 2: components/potentials.py — add WrapSerializer to _AnnotatedCollections

Without serialize_as_any=True, pydantic serializes collections: dict[str, Collection] using the declared Collection base class schema, silently dropping subclass-specific fields (scale_14, cutoff, periodic_potential, etc. from _NonbondedCollection and subclasses). This caused wrong electrostatics energies after JSON roundtrip:

AssertionError: assert 0.5 == 0.8333333333  # scale_14 reset to default
EnergyError: {'Electrostatics': <Quantity(-17.38, 'kilojoule / mole')>}

The fix adds a WrapSerializer to _AnnotatedCollections that calls model_dump_json() on each collection instance directly. Since the call is on the actual instance (e.g. SMIRNOFFElectrostaticsCollection), pydantic uses the full subclass schema and all fields are preserved.

Fixes #1346.

Test results (pydantic 2.13.3, after both fixes, py312amber environment)

612 passed, 119 skipped, 13 xfailed, 8 xpassed in 248.98s

Zero failures. Mypy also passes (Success: no issues found in 85 source files).

[michalkrompiec]

Oops, it turns out I have not run the tests properly locally. Will come back when fixed!

[michalkrompiec]

This version passes the py312amber test on my machine. Can you try running the CI again, please?

[michalkrompiec]

Added new pydantic to the environment and added a test explicitly using pydantic>2.12.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.73%. Comparing base (14cd931) to head (7f80e6b).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1493      +/-   ##
==========================================
- Coverage   94.07%   89.73%   -4.35%     
==========================================
  Files          73       73              
  Lines        6045     6048       +3     
==========================================
- Hits         5687     5427     -260     
- Misses        358      621     +263     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[michalkrompiec]

Openeye tests failing due to lack of license keys - is it because the branch is on my account?