should fix issue #7520, test_nanargmin fails with datetime64 objects

[toddrjen]

I think this will fix the issue #7520. The issue appears to be that older versions of numpy, probably less than 1.7.0, can't handle datetime64 dtypes very well. I have added simple check to see if numpy can handle a standard operation with datetime64 dtypes, and if not it skips those tests.

I can't test this personally since I don't have the older version of numpy.

[jreback]

all datetime64/timesela64 HAVE TO be cast to i8 before operations (and cast back after)

it is an assertion error if this is not the case

[jreback]

you can simply set up a virtual env to test numpy 1.6.1

[toddrjen]

Yes, nanops casts internally. The problem with this bug is with the reference function. All the nanops tests compare to a reference function to make sure the calculations are correct. In this case, it is a numpy function. The current version of the numpy function, and the version that travis tests against, handle datetime64 objects fine. Older versions, however, apparently don't.

So this change just tests the numpy reference function and if it fails, it skips the test.

[jreback]

that's fine then