support decoding to `collections.Counter`

Description

This is something pydantic supports, but not msgspec:

from collections import Counter

c: Counter[str] = Counter()

c["abc"] += 3
c["xyz"] += 1


def pyd() -> None:
    import pydantic
    CounterStr = pydantic.TypeAdapter(Counter[str])
    c_serialized = CounterStr.dump_json(c)
    print(c_serialized)
    c_deserialized = CounterStr.validate_json(c_serialized)
    print(c_deserialized)


def msg() -> None:
    import msgspec
    CounterStr = msgspec.json.Decoder(Counter[str])
    c_serialized = msgspec.json.encode(c)
    print(c_serialized)
    c_deserialized = CounterStr.decode(c_serialized)
    print(c_deserialized)


pyd()
msg()

output:

b'{"abc":3,"xyz":1}'
Counter({'abc': 3, 'xyz': 1})
b'{"abc":3,"xyz":1}'
Traceback (most recent call last):
...
msgspec.ValidationError: Expected `Counter`, got `dict`

Start with the msgspec.json.Decoder(Counter[str]) example and compare its behavior with pydantic's Counter handling. Add support so decoding the encoded mapping returns a Counter with the same counts, and cover the behavior with a regression test for the shown example.