process: move all the optimizations of switchout() into the C code

process: move all the optimizations of switchout() into the C code

This fixes a ton of duplication in the asm code, and in addition makes the
pre-emption cases faster as we know in those cases that the optimizations
never apply.

switchout() becomes a C fnction

platform_switchout() is now the required platform specific stub which shouldn't
need to do any optimizing beyond udata copy elimiations.

Merge a z80fixedbank example of the changes.