D417 - Three more HLE fixes from the same diff: thread join, mutex type, audio init

measured - 2026-08-31

Continuing D416 down the hardware-vs-orbistoun divergence list, each confirmed by the verdict flipping when obSCEne re-runs:

sceKernelCreateSema (018-relational/handle-fits) was on the list and deliberately left alone: its verdict differs because hardware fails it - the console writes past the end of the int - while orbistoun writes a clean four-byte handle. orbistoun is the more correct of the two, and matching the quirk would mean writing a bug.

Total across D416+D417 vs the pre-fix run: 516 -> 522 pass, 12 -> 8 partial, 8 -> 6 skip, fails unchanged, plus the LoadStartModule and mutex measures now matching hardware. Recorded measured.