PYTHON-4663 Fix compatibility with dateutil timezones

[ShaneHarvey]

PYTHON-4663 Fix compatibility with dateutil timezones

Depends on #1811

Here's a quick benchmark to sanity check we're not regressing in performance:

import datetime
import timeit
from bson import DatetimeMS, decode, encode, CodecOptions, DatetimeConversion

MIN = -62135596800000
MAX = 253402300799999
N = 10000
clampopts = CodecOptions(datetime_conversion=DatetimeConversion.DATETIME_CLAMP, tz_aware=True, tzinfo=datetime.timezone.utc)
awareopts = CodecOptions(datetime_conversion=DatetimeConversion.DATETIME, tz_aware=True)
clampawareopts = CodecOptions(datetime_conversion=DatetimeConversion.DATETIME_CLAMP, tz_aware=True)

def decode_in_range_default():
    return decode(data_in)
def decode_in_range_default_aware():
    return decode(data_in, awareopts)
def decode_in_range_default_aware_clamp():
    return decode(data_in, clampawareopts)
def decode_in_range():
    return decode(data_in, codec_options=clampopts)
def decode_out_of_range():
    return decode(data_out, codec_options=clampopts)
def encode_in_range():
    return encode(doc_in)
def encode_out_of_range():
    return encode(doc_out)

doc_in = {"d": [datetime.datetime(1970, 1, 1)] * N}
doc_out = {"d": [DatetimeMS(MIN-1)] * N}
data_in = encode_in_range()
data_out = encode_out_of_range()

if __name__ == '__main__':
    for f in [
        decode_in_range_default, decode_in_range_default_aware, decode_in_range_default_aware_clamp,
        decode_in_range, decode_out_of_range, encode_in_range,
    ]:
        res = timeit.timeit(f"{f.__name__}()", setup=f"from __main__ import {f.__name__}", number=100)
        print(f"{f.__name__}: {res}")

Before:

$ python datetime-bench.py         
decode_in_range_default: 0.06987654202384874
decode_in_range_default_aware: 0.5483650420210324
decode_in_range_default_aware_clamp: 0.6285510419984348
decode_in_range: 0.9283483749604784
decode_out_of_range: 1.639427791989874
encode_in_range: 0.06142729101702571

After:

$ python datetime-bench.py         
decode_in_range_default: 0.06868383300025016
decode_in_range_default_aware: 0.5209614590276033
decode_in_range_default_aware_clamp: 0.5209896659944206
decode_in_range: 0.8219957919791341
decode_out_of_range: 1.5381669579655863
encode_in_range: 0.06304012500913814

In fact we're now 10% faster in decoding aware datetimes (both in and out of range). So it appears is is faster to recompute _min_datetime_ms/_max_datetime_ms in the C extensions than it is to lookup the cached value.

[ShaneHarvey]

Ready for review. I opened PYTHON-4701 for the pypy test failure.

[blink1073]

LGTM!