Simplified doc about tests in Windows, since the script to fix this test issue is not currently working

Added an option in pydotprint* to put less information in the graph when we have gived a name to a variable.

Also got rid of warning in one of these tests

trunk of Theano, it can only be produced by Arnaud optimization (that fuses
scan ops together). Unfortunately due to lack of testing, the optimization
is not yet added into the trunk. The bug is that the scan_op generated by
the gradient uses the same variables as the original scan, and if you try to
fuse these two ops you run into troubles.