Pandas BUG: Faulty DatetimeIndex union

Pandas version checks

• [x] I have checked that this issue has not already been reported.

• [x] I have confirmed this bug exists on the latest version of pandas.

• [x] I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

def test_pandas_datetime_index_union(self):
    """
    Demonstrates a suspected bug in pandas 2.2.3, where unions of DatetimeIndexes (and therefore pd.concats of dataframes with DatetimeIndexes)
    are returning unexpected values.

    My actual usecase (concatenating two dataframes with these DatetimeIndexes, from which I extracted these date ranges) works in
    pandas 1.5.3, but not 2.2.3.

    Interestingly, this passes in both versions if you change the dtype to datetime64[ns].

    """

    dti1 = DatetimeIndex(
        ['2021-10-05 17:30:00', 
         '2021-10-05 18:00:00', 
         '2021-10-05 18:30:00', 
         '2021-10-05 19:00:00', 
         '2021-10-05 19:30:00'],
        dtype='datetime64[us]', name='DATETIME', freq='30min'
    )
    dti2 = DatetimeIndex(
        ['2021-10-05 17:30:00', 
         '2021-10-05 18:00:00',  
         '2021-10-05 18:30:00', 
         '2021-10-05 19:00:00', 
         '2021-10-05 19:30:00', 
         '2021-10-05 20:00:00'], # <-- Extra datetime
        dtype='datetime64[us]', name='DATETIME', freq='30min'
    )

    union = set(dti1.union(dti2))
    expected = set(dti1) | set(dti2)
    print(f"{union=}")
    print(f"{expected=}")

    assert len(union) == len(expected), "Should have all the rows from the concatenated dataframes"


def test_range_index_equality(self):
    """ This (presumably) faulty equality check appears to be the root cause of the datetimeindex union bug above
    Note that the two stop values are different, so the RangeIndexes should not be equal.
    Interestingly, this fails in both pandas 1.5.3 and 2.2.3.
    """
    a = RangeIndex(start=1633455000000000, stop=1635262200000000, step=1800000000000)
    b = RangeIndex(start=1633455000000000, stop=1635264000000000, step=1800000000000)
    assert not a.equals(b)

Issue Description

These tests above (details in the function doc) demonstrate the issue and what I think is the root cause. Basically we get back what appears to be an incorrect result when taking the union of two DatetimeIndexes with different ranges.

I traced this as far as the RangeIndex equality check in the second test, which appears to be faulty, returning True for two different stop values.

Expected Behavior

Out from first test should be (as in pandas 1.5.3):

union={Timestamp('2021-10-05 18:30:00'), Timestamp('2021-10-05 19:00:00'), Timestamp('2021-10-05 17:30:00'), Timestamp('2021-10-05 20:00:00'), Timestamp('2021-10-05 19:30:00'), Timestamp('2021-10-05 18:00:00')}

expected={Timestamp('2021-10-05 18:30:00'), Timestamp('2021-10-05 19:00:00'), Timestamp('2021-10-05 17:30:00'), Timestamp('2021-10-05 20:00:00'), Timestamp('2021-10-05 19:30:00'), Timestamp('2021-10-05 18:00:00')}

But the actual output in pandas 2.2.3 is (incorrectly):

union={Timestamp('2021-10-05 17:30:00'), Timestamp('2021-10-26 13:30:00')}

expected={Timestamp('2021-10-05 19:30:00'), Timestamp('2021-10-05 20:00:00'), Timestamp('2021-10-05 17:30:00'), Timestamp('2021-10-05 19:00:00'), Timestamp('2021-10-05 18:00:00'), Timestamp('2021-10-05 18:30:00')}

Installed Versions