Speed up postgrest JSON validation

[snoopuppy582]

Fixes #1486

Summary

• switch postgrest.types.JSONAdapter to Pydantic's optimized JsonValue schema
• keep the public JSON type alias unchanged for request typing
• add regression coverage that guards the JsonValue adapter path and nested JSON byte validation

Local validation

• Before fix benchmark from the issue payload: JSONAdapter.validate_json(payload) averaged 44.78 ms; TypeAdapter(JsonValue) averaged 3.93 ms
• After fix benchmark from the issue payload: JSONAdapter.validate_json(payload) averaged 3.56 ms; TypeAdapter(JsonValue) averaged 3.74 ms
• uv run --project src/postgrest pytest src/postgrest/tests/test_types.py -q
• uv run --project src/postgrest pytest src/postgrest/tests/test_types.py src/postgrest/tests/test_utils.py src/postgrest/tests/_sync/test_request_builder.py src/postgrest/tests/_async/test_request_builder.py -q
• uv run --project src/postgrest ruff check src/postgrest/src/postgrest/types.py src/postgrest/tests/test_types.py

Notes

• I tried the full src/postgrest/tests suite locally, but it did not finish before the 4-minute timeout in my Windows environment. The targeted request-builder/type tests above completed successfully.
• uv run --locked currently reports that the existing workspace lockfile needs an update, so the PR intentionally does not include lockfile changes.

[AmaLS367]

Thanks for this PR — the performance improvement is impressive and the change is clean. The core fix looks correct.

One suggestion on the test test_json_adapter_uses_pydantic_json_value_schema:

assert JSONAdapter.core_schema["schema"]["schema_ref"].startswith(
    "pydantic.types.JsonValue:"
)

The intent is good (guarding the fast path), but the nested ["schema"]["schema_ref"] path and the hardcoded "pydantic.types.JsonValue:" string tie the test to Pydantic's internal schema structure. A more resilient alternative would be to compare against a reference adapter:

def test_json_adapter_schema_matches_json_value():
    from pydantic.types import JsonValue as PydanticJsonValue
    from pydantic import TypeAdapter
    assert JSONAdapter.core_schema == TypeAdapter(PydanticJsonValue).core_schema

This guards the same regression (someone reverting to the slow TypeAliasType path) without hardcoding internal schema details. Just a suggestion though — not a blocker.

[snoopuppy582]

Thanks for the suggestion. I updated the test to compare JSONAdapter.core_schema against TypeAdapter(JsonValue).core_schema instead of asserting the nested Pydantic schema ref string.

Verified locally:

• uv run pytest tests/test_types.py -q -> 2 passed