BUG: fixed value_counts with datetime64[ns], GH 3002

[jreback]

Fixes issue #3002

• Fixed for datetime64[ns]
• Fixed for timedelta64[ns](though the returned index
  is actually int64, which is wrong, should make a TimeDeltaIndex class....someday)
• Fixed NaT display in a DateTimeIndex (was displaying funny)

0   2010-01-01 00:00:00
1   2010-01-01 00:00:00
2   2009-01-01 00:00:00
3   2008-09-09 00:00:00
Name: dt, dtype: datetime64[ns]

2010-01-01    2
2008-09-09    1
2009-01-01    1
dtype: int64

<class 'pandas.tseries.index.DatetimeIndex'>
[2010-01-01 00:00:00, ..., 2009-01-01 00:00:00]
Length: 3, Freq: None, Timezone: None

WIth NaTs

0   2010-01-01 00:00:00
1   2010-01-01 00:00:00
2   2009-01-01 00:00:00
3   2008-09-09 00:00:00
4                   NaT
dtype: datetime64[ns]

2010-01-01    2
2009-01-01    1
NaT           1
2008-09-09    1
dtype: int64

<class 'pandas.tseries.index.DatetimeIndex'>
[2010-01-01 00:00:00, ..., 2008-09-09 00:00:00]
Length: 4, Freq: None, Timezone: None