Bug with the GMM parallel implementation

Created by: tiagofrepereira2012

I found a bug in the GMM parallel implementation.

The number of statistics files that is written in the e-step (KMeans and M. Likelihood) are not the same amount of statistics read in the m-step.

Basically the check here https://github.com/bioidiap/bob.bio.gmm/blob/master/bob/bio/gmm/tools/gmm.py#L40 must match with the check here https://github.com/bioidiap/bob.bio.gmm/blob/master/bob/bio/gmm/tools/gmm.py#L124

AND

the check here: https://github.com/bioidiap/bob.bio.gmm/blob/master/bob/bio/gmm/tools/gmm.py#L196 must match with the check here https://github.com/bioidiap/bob.bio.gmm/blob/master/bob/bio/gmm/tools/gmm.py#L250

The crash happens only when you have few training files in the training set. I'm opening the issue just to register. I will push the patch soon. @siebenkopf , can I release a new version on pypi after this?