Jit_Integer: Optimize rlwimix

[merryhime]

Last PR for a while.

The main purpose of this PR was the final commit, wherein we implement cases where mask is 0xFF or 0xFFFF with partial register writes.

A few additional commits for other things noticed along the way, these can be split into separate PRs if desired.

Sidenotes for future optimziations:

• Using AH/BH/CH/DH is unlikely to be sufficiently profitable due to encoding limitations.
• Using pinsrb is unlikely to be profitable due to the cost of register moves.