BUG: casting dt64 in DataFrame.reindex

[jbrockmendel]

import numpy as np
import pandas as pd

mi = pd.MultiIndex.from_product([list("ABCDE"), range(2)])
dti = pd.date_range("2016-01-01", periods=10)
ser = pd.Series(dti, index=mi)
df = ser.unstack()

index = df.index.append(pd.Index([1]))
columns = df.columns.append(pd.Index(["foo"]))
axes = {"index": index, "columns": columns}

fv = np.timedelta64("NaT", "ns")

res = df.reindex(fill_value=fv, **axes)

>>> res
>>> res
                     0                    1 foo
A  1451606400000000000  1451692800000000000 NaT
B  1451779200000000000  1451865600000000000 NaT
C  1451952000000000000  1452038400000000000 NaT
D  1452124800000000000  1452211200000000000 NaT
E  1452297600000000000  1452384000000000000 NaT
1                  NaT                  NaT NaT

The underlying problem here is numpy/numpy#12550