use unmasked data

[bjlittle]

Surprisingly using masked arrays can be of the order 4-5 times slower (for iris-agg-regrid) than a regular numpy.ndarray.

Therefore, if there are no masked data in a masked array, then use the underlying numpy.ndarray instead.

[coveralls]

Coverage increased (+4.5%) to 95.222% when pulling a777fcd on bjlittle:use-unmasked-data into e4bdc74 on SciTools-incubator:master.

[coveralls]

Coverage decreased (-0.6%) to 90.102% when pulling 153e32b on bjlittle:use-unmasked-data into e4bdc74 on SciTools-incubator:master.