JIT: Don't needlessly flush register state in merged integer+branch code

[FioraAeterna]

The actual branch code doesn't flush the register state, so it doesn't make
much sense for the merged optimization to do so either.

Not tested for performance yet, but it surely can't be /slower/.

[FioraAeterna]

Also if anyone knows why it was done this way (or if there's a good reason) I'd love to know; I encountered this while trying to write merged float cmp/branch code and found it was slower than master when flushing the register cache, and then noticed that the integer merged branch code was doing the same

[Sonicadvance1]

Right. I missed these with the whole conditional branch + not dumping register cache state a few months ago.

Thanks for catching these!