Unstructure subclassed built-in collections like their built-ins - #824

Fixes #287.

Subclasses of built-in collections were not un/structured like their built-ins: converter.unstructure(MyList([Item(1)])) returned the raw Item objects, and a field annotated MyList[Item] was passed through unchanged too, while list[Item] worked.

Two gaps in the type predicates in cattrs._compat:

1. The collection predicates (is_sequence, is_mutable_sequence, is_tuple, is_deque, is_mutable_set, is_frozenset, is_mapping) only recognized typing._GenericAlias. Subscripting a user class produces a PEP 585 types.GenericAlias instead, so MyList[Item] never matched. These predicates now accept both alias types (as is_generic already did).
2. Unlike is_mapping, the sequence and set predicates had no clause for bare subclasses, so dispatching on a bare MyList class (for example when unstructuring an instance directly) fell through to the identity fallback. is_mutable_sequence, is_mutable_set and is_frozenset now also match bare subclasses of list/deque, set and frozenset respectively, mirroring the existing is_mapping clause.

Class-registered hooks (for example for str, bytes or Path subclasses) still take precedence, since singledispatch runs before these predicates, and attrs classes are still matched earlier by the has predicate.

Tests:

• Added test_unstructure_subclassed_list and test_unstructure_subclassed_sets to tests/test_baseconverter.py, covering bare instances, bare and subscripted field annotations, and a structuring round-trip.
• Regression runs: tests/test_baseconverter.py, tests/test_gen_dict.py, tests/test_dicts.py, tests/test_function_dispatch.py, tests/test_multistrategy_dispatch.py, tests/test_tuples.py, tests/test_structure.py, tests/test_unstructure.py, tests/test_unstructure_collections.py, tests/test_cols.py, tests/test_defaultdicts.py, tests/test_typeddicts.py — all passed (257 tests).
• ruff check and ruff format --check on the changed files: pass.

A changelog entry is included in HISTORY.md.