Potential refactor of crypto operations

[0xkanekiken]

According to @bobbinth, we could make mtree_set a bit more efficient by changing semantics of MRUPDATE from:

[V, d, i, R, NV] -> [NR, d, i, R, NV]

to

[V, d, i, R, NV] -> [V, d, i, NR, NV]

This should shave off 3 VM cycles from mtree_set and get it down to 11 cycles, but:

• It would complicate handling of constraints for this operation since we change items at odd stack positions.
• It would make mtree_cwm operation quite a bit less efficient.

[bobbinth]

Superseded by #654.