ENH: using sklearn.utils for creating RNGs and generating minibatches

[dohmatob]

Spotlight already depends on scikit learn (sklearn), so we could delegate even more stuff to the later.

[maciejkula]

Looks good, thanks for the contribution!

Could you fix the flake8 issue please?

[dohmatob]

Fixed the flake issue. Now there is a mystical failure which seems to be affecting other open PRs too.

UnsatisfiableError: The following specifications were found to be in conflict:
  - chardet -> python[version='>=3.6,<3.7.0a0']
  - python=2.7
Use "conda info <package>" to see the dependencies for each package.

[maciejkula]

Yes, the dependencies seem to have moved in an annoying way. I'll do my best to resolve this.

[maciejkula]

Could you please rebase on top of newest master?

[maciejkula]

Can you rebase please? Otherwise I'll close this and re-open the PR with rebased code.