Fix FilterSchema list parsing for PEP 695 type aliases (#1720)

[Vansh-Sharma27]

Summary

Fixes #1720.

When a PEP 695 type alias for a list[...] type is used in a FilterSchema field, the query parameter fails to parse:

{"detail": [{"type": "list_type", "loc": ["query", "name"], "msg": "Input should be a valid list"}]}

Minimal repro from the issue:

type name_list = Annotated[list[str] | None, FilterLookup("name__in")]

class BookFilter(FilterSchema):
    name: name_list = Field(default=None)

Root cause

type X = ... produces a typing.TypeAliasType whose real annotation lives in __value__, and get_origin() on it returns None. Pydantic stores the alias verbatim as the field annotation, so:

1. is_collection_type() in ninja/signature/details.py never sees the underlying list, so the field is not registered as a collection and repeated query params are not collected into a list.
2. FilterSchema._get_filter_lookup() reads field_info.metadata, which is empty for an aliased field, so a FilterLookup defined inside the alias was silently dropped.

Point 2 matters: fixing only the parsing turns the loud 422 into a silently wrong lookup (name instead of name__in), so both paths needed handling.

Changes

• Add TYPE_ALIAS_TYPES to ninja/compatibility/util.py (covers both the stdlib 3.12+ class and the typing_extensions backport).
• is_collection_type() now unwraps TypeAliasType (and a contained Annotated) before checking the origin.
• _get_filter_lookup() now also reads FilterLookup metadata from inside the alias.

Non-list aliases and plain Annotated fields are unchanged.

Tests

Added three tests to tests/test_filter_schema.py:

• test_filter_lookup_from_list_type_alias — the inner FilterLookup is applied.
• test_list_type_alias_query_parsing — repeated query params parse into a list (this is the exact 422 from the issue).
• test_non_list_type_alias_still_works — regression guard for scalar aliases.

The tests build the alias via TypeAliasType(...) (the same runtime object the type statement creates) so the file still imports on Python < 3.12. I confirmed the two bug-specific tests fail on master and pass with the fix.

pytest, ruff format --check, ruff check, and mypy ninja all pass.