The formula before was wrong, it does not work when inputs are matrices,
because it depends on which axis y is computed.  I believe this formulation
is much better.

Before only the gradient of inc_subtensor was ok ( and it also was the only
one checked)

numpy.all doesn't check dtype and shape

Rationale: its name shadowed the scipy package, and it was not successful in
simplifying other modules' code that had to do scipy version checks. (Which is
only two places anyway).

The test checks the depricated return_steps

This will help during debugging

After talking to Pascal, I decided to leave until in scan_module and not to
export it to theano.until. So I had to fix the tests as well for this.

I needed to do this when debugging, because I wanted to know both of their
values, but there was nothing wrong with the test, and with or without this
change it is the same. I just wanted to stay in a more clear form ( so
people don't get confused by which f1 we talking about)

return_steps has been depricated, and store_steps as well (it actually not
used anymore)