BUG: conversion from np.datetime64 has odd numpy buglet

[jreback]

see #3414

The resulting dtype is wrong (should be datetime64[ns]),
and should either raise on dtype='datetime64' or deal with it in Series

In [1]: dates = [
   ...:     np.datetime64(datetime.date(2013, 1, 1)),
   ...:     np.datetime64(datetime.date(2013, 1, 2)),
   ...:     np.datetime64(datetime.date(2013, 1, 3)),
   ...: ]

In [2]: s = pd.Series(dates, dtype='datetime64')

In [3]: s.ix[0] = np.nan

In [4]: print s
0                   NaT
1   2013-01-02 00:00:00
2   2013-01-03 00:00:00
dtype: datetime64[us]

This should force conversion on the np.datetime64 (regardless of there dtype)

In [22]: pd.Series(dates)
Out[22]: 
0    2013-01-01 00:00:00
1    2013-01-02 00:00:00
2    2013-01-03 00:00:00
dtype: object

Clearly incorrect interpret of np.datetime64 dtypes as well (its interpreting them as M8[ns], not converting
to)

In [25]: pd.Series(dates,dtype='M8[ns]')
Out[25]: 
0   1970-01-16 16:56:38.400000
1   1970-01-16 16:58:04.800000
2   1970-01-16 16:59:31.200000
dtype: datetime64[ns]

[jreback]

closed by #3516