loffset has no effect when passing in a numyp.timedelta64

[MarcMassar]

From mailing-list: [https://groups.google.com/forum/#!topic/pydata/CDnF9cNR2ho]
When doing a re-sample, loffset has no effect if a numpy.timedelta64 is passed in:

In [1]: rng = pd.date_range('1/1/2012', periods=4, freq='S')

In [2]: ts = pd.Series(np.random.randint(0, 500, len(rng)), index=rng)

In [3]: ts.resample('2S')
Out[3]:
2012-01-01 00:00:00    396.0
2012-01-01 00:00:02    214.5
Freq: 2S, dtype: float64

In [4]: import datetime

In [5]: ts.resample('2S',loffset=datetime.timedelta(seconds=10))
Out[5]:
2012-01-01 00:00:10    396.0
2012-01-01 00:00:12    214.5
dtype: float64

In [6]: ts.resample('2S',loffset=np.timedelta64(10,'s'))
Out[6]:
2012-01-01 00:00:00    396.0
2012-01-01 00:00:02    214.5
Freq: 2S, dtype: float64

[jreback]

would welcome a pull-request for this
also should raise on an invalid loffset