A larry of all NaNs does not return NaN for some methods

[kwgoodman]

The following methods do not return NaN when the larry contains all NaNs: prod, cumprod, cumsum.

Current behavior:

>> la.larry([la.nan, la.nan]).prod()
   1.0
>> la.larry([la.nan, la.nan]).cumprod(axis=0)
label_0
    0
    1
x
array([ 1., 1.])
>> la.larry([la.nan, la.nan]).cumsum(axis=0)
label_0
    0
    1
x
array([ 0., 0.])

Desired behavior:

>> la.larry([la.nan, la.nan]).prod()
   NaN
>> la.larry([la.nan, la.nan]).cumprod(axis=0)
label_0
    0
    1
x
array([ NaN, NaN])
>> la.larry([la.nan, la.nan]).cumsum(axis=0)
label_0
    0
    1
x
array([ NaN, NaN])