BUG: Index with integer data and datetime64[ns, tz] dtype does not localize correctly

[mroeschke]

In [1]: pd.__version__
Out[1]: '0.23.0rc2+16.gccf4b96.dirty'

In [2]: val = [pd.Timestamp('2018-01-01', tz='US/Pacific').value]

In [3]: pd.Index(val, dtype='datetime64[ns, US/Pacific]')
Out[3]: DatetimeIndex(['2018-01-01 08:00:00-08:00'], dtype='datetime64[ns, US/Pacific]', freq=None)

The localization appears to localize directly to the timezone instead of localizing first to UTC and therefore does not roundtrip correctly from the timestamp value.

#20956 can be simplified once this is fixed.

Expected:

Out[3]: DatetimeIndex(['2018-01-01 00:00:00-08:00'], dtype='datetime64[ns, US/Pacific]', freq=None)

[jreback]

hmm I think there is an open issue about this already, if you'd have a look