fast tensor canonicalization

[mario]

I have made a pull request https://github.com/sympy/sympy/pull/1326
on the canonicalization of tensors with dummy indices;
it is a nice application of the Schreier-Sims algorithm.

It is faster than xperm.c of the XPerm package in the few examples
I tried (by a wide margin for very long tensors);
I have put in the tests a couple of
examples from test_xperm.c in cadabra, one of which runs in 0.1s
and hangs with xperm.c

[krastano...@gmail.com]

Thanks for the addition. For the moment I only skimmed over the code
and focused on the docstrings. I found it hard to understand. Could
you try to explain a bit more (perhaps the pull request page is better
suited for the discussion).