BUG: adding np.timedelta64 to DatetimeIndex with tz outputs incorrect

[sinhrks]

>>> idx = pd.date_range(start='2010-11-02 01:00:00', periods=3, tz='US/Pacific', freq='1H')

>>> idx + offsets.Hour(3)
[2010-11-02 04:00:00-07:00, ..., 2010-11-02 06:00:00-07:00]
Length: 3, Freq: H, Timezone: US/Pacific

>>> idx + datetime.timedelta(hours=3)
[2010-11-02 04:00:00-07:00, ..., 2010-11-02 06:00:00-07:00]
Length: 3, Freq: H, Timezone: US/Pacific

>>> idx + np.timedelta64(3, 'h')          # incorrect
[2010-11-02 11:00:00-07:00, ..., 2010-11-02 13:00:00-07:00]
Length: 3, Freq: H, Timezone: US/Pacific

[jreback]

looks good

[jreback]

trivial, but can you add the issue number (in this case PR number), as a comment near tests, easier reference for later

[sinhrks]

OK, I've added a comment and rebased. There seems to be unnecessary line remains in release.rst thus removed it.