Rescale with featurewise normalization bug fix

[hadipash]

Summary

Featurewise centering, normalization and ZCA whitening work incorrectly with the rescale parameter. This PR fixes the bug.

Related Issues

100

PR Overview

• [y] This PR requires new unit tests [y/n] (make sure tests are included)
• [n] This PR requires to update the documentation [y/n] (make sure the docs are up-to-date)
• [y] This PR is backwards compatible [y/n]
• [n] This PR changes the current API [y/n] (all API changes need to be approved by fchollet)

[hadipash]

Thanks @rragundez for your help! Few comments on the latest commit. I followed your suggestions on enhancing test cases but in the case of featurewise_std_normalization it's impossible to achieve 1e-6 precision because on the lines 722 and 967 of image_data_generator.py X is divided by the standard deviation + 1e-6 (x /= (self.std + 1e-6)), so the precision is smaller. Therefore, I set the level to 1e-5.

Additionally, I added dtype argument for iterators initialization so that data can be treated as other types rather than only float32 (as explained above).

Finally, I introduced a better test case for ZCA whitening (also explained above).