la.afunc.nans: wrong output for str and object dtype

[kwgoodman]

nans (similar to numpy.ones and numpy.zeros) should fill with '' when dtype is str and None when dtype is object. A better error message would be useful for when dtype is int. Current behavior:

>> nans(2)
   array([ NaN, NaN]) # <--- Good
>> nans(2, dtype=str)

array(['n', 'n'],
      dtype='|S1') # <--- Bad, fill with ''
>> nans(2, dtype=object)
   array([nan, nan], dtype=object) # <--- Bad, fill with None
>> nans(2, dtype=int)
[...]
ValueError: cannot convert float NaN to integer # <--- Need better error msg