BUG: inferred Timestamp unit with dateutil paths

[jbrockmendel]

>>> pd.Timestamp("10/27/23 14:19:01.000").unit
's'

>>> pd.Timestamp("01-01-2013T00:00:00.000000000+0000").unit
's'

In dateutil_parse we have

    if reso == "microsecond":
        if repl["microsecond"] == 0:
            reso = "second"
        elif repl["microsecond"] % 1000 == 0:
            reso = "millisecond"

which ends up ignoring the trailing zeros. We end up with an incorrectly low-resolution unit.

Surfaced while implementing #55564.