fix(reassembly): Deferred(Generic[_KT]) still fails to pickle under protocol 0/1 on GraalPy #2433

Deferred(Generic[_KT]) still fails to pickle under protocol 0/1 on GraalPy, after #2424 explicitly added __getstate__/__setstate__ to fix exactly this for CPython. Found via the experimental: main push runs Discussion thread (run 40909721813, commit 857bd522e), Python graalpy-25.4 (experimental) leg, non-blocking but a real defect:

tests/foundation/reassembly/test_deferred_pickle_2423_unit.py::DeferredPickleTests::test_protocol_0_round_trips
tests/foundation/reassembly/test_deferred_pickle_2423_unit.py::DeferredPickleTests::test_protocol_1_round_trips
tests/foundation/reassembly/test_deferred_pickle_2423_unit.py::DeferredPickleSlotSpellingTests::test_bare_string_slot_round_trips_under_every_protocol (protocol 0/1)
tests/foundation/reassembly/test_deferred_pickle_2423_unit.py::DeferredPickleSlotSpellingTests::test_mangled_slot_round_trips_under_every_protocol (protocol 0/1)

All fail with:

state = base(self)
TypeError: typing.Generic() takes no arguments
/opt/hostedtoolcache/GraalPy/25.4.4/x64/lib/python3.13/copyreg.py:71: TypeError

Root cause: for pickle protocol < 2, copyreg._reduce_ex walks cls.__mro__ before consulting __getstate__, looking for the first base that is not a heap type (or whose __new__ is bound to itself) to compute a base-object state via base(self). On CPython that walk passes through Generic and lands on object (so state = None). On GraalPy, Generic.__new__ apparently satisfies the walk's stopping condition, so base resolves to Generic itself, and Generic(self) raises — #2424's __getstate__/__setstate__ only supplies the dict half of the reduction, not this base-object half, so it doesn't help.

Likely fix: implement Deferred.__reduce_ex__ (or __reduce__) directly for protocol 0/1, bypassing copyreg._reduce_ex's MRO walk entirely, rather than relying on __getstate__ alone — or drop the Generic[_KT] base if it's not load-bearing at runtime (it was added by #2417/#2418 for static typing only, Deferred.__subclasses__/runtime behavior doesn't depend on it).

Non-blocking (experimental GraalPy leg), but real — #2423/#2424's own test file is what's catching it.