fix: ModelSchema doesn't apply alias_generator to foreign key fields

[shivanij1203]

Summary

When a ModelSchema sets an alias_generator (e.g. to_camel), regular fields are aliased but a foreign key's generated _id field is not — it stays snake_case while everything else is transformed:

{
  "id": "...",
  "model_a_id": ...,        // stays snake_case
  "myOtherProperty": "..."  // camelCased
}

The reason is that relation fields set their alias explicitly (to the _id attname) in get_schema_field. In pydantic v2 an explicit alias takes precedence over the model's alias_generator, so the generator never runs on the id field.

The fix threads the schema's alias_generator (from the base class' model_config) into relation-field construction and applies it to the attname, so the id field is aliased consistently with the rest of the schema. When no generator is configured, behavior is unchanged (model_a_id).

Fixes #1691

Changes

• ninja/orm/fields.py: get_schema_field accepts an optional alias_generator and applies it to a relation's _id attname (supports a plain callable and AliasGenerator).
• ninja/orm/factory.py: read alias_generator from the base class' model_config and pass it through when building fields.
• tests/test_orm_relations.py: regression test asserting a FK's id field is camelCased like other fields.

Testing

• pytest tests/test_orm_relations.py tests/test_orm_metaclass.py tests/test_models.py passes.
• The new test fails against main (the FK alias is target_id instead of targetId) and passes with the fix.
• ruff format --check, ruff check, and mypy ninja are all clean.

[mbaragiola]

Blocker

• ninja/orm/fields.py:122: AliasGenerator supports independent alias, validation_alias, and serialization_alias callbacks, but _apply_alias_generator() consults only .alias. If .alias is unset—for example, AliasGenerator(validation_alias=to_camel, serialization_alias=to_pascal)—the foreign-key field keeps target_id for both directions while ordinary fields receive their configured aliases. Please generate and assign the validation and serialization aliases independently, and add runtime validation plus model_dump(by_alias=True) coverage rather than checking only field metadata.

Positives

• The common callable alias_generator=to_camel case is addressed.
• The test compares the foreign-key alias with an ordinary generated field.

Verification

I reviewed the factory/field path and Pydantic’s AliasGenerator.generate_aliases() behavior. GitHub currently reports no checks for this head; I did not run the project suite locally.

Reviewed by Hermes Agent using Codex gpt-5.6-sol.

[shivanij1203]

Good catch, thanks. Reworked this so the _id field goes through pydantic's own AliasGenerator.generate_aliases() instead of only reading .alias. alias, validation_alias, and serialization_alias are now generated and assigned independently, and each direction the generator leaves unset falls back to the _id attname (so a single-callback generator like alias_generator=to_camel still covers both directions, unchanged from before).

For the AliasGenerator(validation_alias=to_camel, serialization_alias=to_pascal) case you flagged, the author_id field now gets authorId for validation and AuthorId for serialization rather than staying author_id in both.

Also swapped the test from metadata-only checks to a behavioral one: it validates via the camelCase validation alias (model_validate) and asserts the PascalCase serialization alias comes back out of model_dump(by_alias=True), alongside a plain field for comparison. mypy ninja and ruff (format + check) are clean.