cv_compute ml_optimise

[ordinarymath]

As in title make it use cv_compute instead of EVAL.

[myreen]

Keep in mind that there is always a cost of going in and going out of cv_compute. I suspect the passes in ml_optimise are so straightforward that they are faster as EVAL calls rather than cv_eval calls.

[ordinarymath]

Hmm maybe optimizing it would be better. The current version of ml_optimise allocates a bunch of completely new terms when rewriting that makes it fairly slow. e.g. (BOTTOM_UP_OPT f (Lit v) = f (Lit v)) /\ creates a new Lit term