I encountered a typing issue when using grpc.aio and figured I would write a PR to solve it at source.

Two related fixes to the async (grpc.aio) metadata types:

1. AioRpcError.trailing_metadata() inherited the synchronous
   RpcError.trailing_metadata() -> tuple[_Metadatum, ...], but at runtime returns a
   grpc.aio.Metadata
   (_call.py).
   Added the override.

2. grpc.aio.Metadata was modelled as a Mapping, but the runtime class is a
   Collection that iterates (key, value) tuples
   (_metadata.py).
   Re-based on Collection[_MetadatumType] with an item-yielding __iter__; keyed
   access (m[key], m.get(...)) is unchanged.

Together these let for key, value in err.trailing_metadata() type-check — it
previously errored with "_Metadatum is not iterable".

import grpc.aio

def reason(err: grpc.aio.AioRpcError) -> None:
    for key, value in err.trailing_metadata():  # before: error "_Metadatum is not iterable"
...

>>> import grpc
>>> from grpc.aio import Metadata, AioRpcError
>>> m = Metadata(("x-request-id", "abc"))
>>> [type(x).__name__ for x in m]            # iterates items, not keys
['tuple']
>>> err = AioRpcError(grpc.StatusCode.NOT_FOUND, Metadata(), m)
>>> type(err.trailing_metadata()).__name__   # not tuple[_Metadatum, ...]
'Metadata'

Includes a regression test. runtests.py --run-stubtest stubs/grpcio passes
(mypy, pyright incl. stricter, test cases, stubtest).

Please let me know if I've overlooked something in the contribution process.