Timedelta fix

[jreback]

b4ee36f should fix it up

my previous was failing because np.array([pd.NaT]]) is object dtype so isnull fails! (which is a separate bug I think)

[danielballan]

One error left:

ERROR: test_timedelta_ops_with_missing_values (pandas.tseries.tests.test_timedeltas.TestTimedeltas)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/dallan/pandas-danielballan/pandas/tseries/tests/test_timedeltas.py", line 247, in test_timedelta_ops_with_missing_values
    actual = s1 + timedelta_NaT
  File "/home/dallan/pandas-danielballan/pandas/core/ops.py", line 473, in wrapper
    time_converted = _TimeOp.maybe_convert_for_time_op(left, right, name)
  File "/home/dallan/pandas-danielballan/pandas/core/ops.py", line 440, in maybe_convert_for_time_op
    return cls(left, right, name)
  File "/home/dallan/pandas-danielballan/pandas/core/ops.py", line 258, in __init__
    rvalues = self._convert_to_array(right, name=name, other=lvalues)
  File "/home/dallan/pandas-danielballan/pandas/core/ops.py", line 375, in _convert_to_array
    " operation".format(pa.array(values).dtype))
TypeError: incompatible type [timedelta64[ns]] for a datetime/timedelta operation

----------------------------------------------------------------------
Ran 829 tests in 43.246s

FAILED (SKIP=16, errors=1)