TF Batch Normalization should use FusedBatchNorm when axis = -1

[yanboliang]

Summary

When applying BatchNormalization to Conv2D, TF backend try to use FusedBatchNorm for optimization(when axis=1 and axis=3), we should cover their equivalent condition(axis=-3 and axis=-1) as well. Since the default value for axis is -1, it's common case that users don't set axis explicitly, then they won't get optimized code route before this fix.

Related Issues

PR Overview

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