Schedule Hopper mma instruction (#3278)

Stacked on #3320 

This PR:
* Schedules the MMA instruction result for the
HopperMultiMatmulScheduler.
* Removes some unused methods that are no longer necessary.
* Checks that there is "no prologue". Specifically, that we have `gmem
-LoadStoreOp-> smem -MmaOp->`. This can currently not be done unless we
create the MmaOp at definition using `fusedMultiplySum` (see #1628).
* Checks that MmaOp output has logical order MNK. If not then a
root->logical reorder should have been created at definition. (maybe
this should be made easier as an option in `fusedMultiplySum`).

This PR does not schedule split-K or TMA stores of the output.

---------

Co-authored-by: Ryan Spring <[email protected]>