use the new DebugMode machanism to disable error when an stability optimization remove nan for the local_{mul,div}_switch_sink optimization. Updated test to use it.

Implemented a new system that allow an optimization to specify a values_eq_approx to use for a variable.

This is usefull to make DebugMode accept that a stabilization optimization remove inf or nan from the graph.

in the stability optimization of the grad of log(erfc(x)) accept more equivalent type of graph and test that the opt are done in the stabilization phase.

add a STABILIZE mode and a stabilize optimizer. They allow to test that stabilization optimization happen in the stabilize phase. Not always the case now.

rename a variable and make the stability optimizer of the grad of log(erfc(x)) happen directly with the canonicalized graph

BUG FIX: the stabilization optimization of the grad of log(erfc(x)) was applied too frequently! That test for that bug

make TensorType.values_eq_approx handle correctly case when their is a mix of inf and nan. When their is nan, we want to accept values that have the same inf.

Added printing of output names for apply nodes in debugprint, such that if you manually assign a name to the outputs (variables), the name will be printed in the debugprint output.

- Code reviewed by Olivier B., thanks!
- Lots of comments added
- Unused code removed
- Looping in Elemwise is now done so the contiguous dimension of the output is
  looped over in the inner-most loop. It does not change anything in the
  non-inplace case (the output was C-contiguous), but can lead to really
  better performances (at least 7x) in the inplace case.

Added flags DebugMode.allow_remove_inf and TensorType.value_eq_approx( allow_remove_inf=False) parameter. When True, we allow an optimization to replace an inf to some other value. Usefull to test stability optimization.