Use divide method (and fix a couple errors that uncovered)

[mwcraig]

The main change was intended to fix #92, but it turns out that running the tests of `flat_correct`` with perfectly uniform flats was masking a couple of errors.

One was that the arithmetic methods return a new array, but flat correct was returning the original array. The other was that it isn't always the case that flat_data.data.mean() == ccd_data.data.mean() after flat correction, because the mean of a ratio isn't the ratio of the means (though this had me going in circles for about 15 minutes).

The tests were also change to almost_equal-type tests.

[coveralls]

Coverage decreased (-0.01%) when pulling c123252 on mwcraig:fix-divide into c2de563 on astropy:master.

[coveralls]

Coverage decreased (-0.01%) when pulling 80bf577 on mwcraig:fix-divide into c2de563 on astropy:master.

[crawfordsm]

Instead of copying the data, multiply flat_corrected by the normalization value so a copy doesn't need to be created.

However, a copy (or other similar type solution ) for when min value is passed in.