New version of ae_op_compat is now proved using the previous result.

This reverts commit 19ecefff.

We did not know how to do without repeating the eqType canonical structure for Ring (it already exists for AbelianGroup).

finite_element.v is still in progress.

Canonical structure is OK, but lemmas addr* are not found!

TODO: the same with rintType (math-comp) -> Ring (Coquelicot).

Make matrices from math-comp an AbelianGroup from Coquelicot.
But it should be more satisfying to use the more generic result above.

WIP: make matrices a ModuleSpace (the ringType K should be seen as a Ring).

re define sigma without 'Hom.