Deferred consolidation: 5-arch setjmp asm + thread core (residual)

[cataggar]

Refs #17

MIGRATED:

• setjmp/longjmp/sigsetjmp global asm in lib/c/setjmp.zig for arm/thumb (arm/armeb/thumb/thumbeb), riscv32/riscv64, powerpc/powerpc64 (powerpc/powerpcle/powerpc64/powerpc64le), and s390x.
• Commented the matching musl setjmp/sigsetjmp asm entries in src/libs/musl.zig.

NOT LANDED / RESIDUAL:

• Cancellation core stays musl C/asm: pthread_cancel.c, __syscall_cp.c, per-arch syscall_cp.s, and SIGCANCEL wiring. This remains the gate for nanosleep/wait*/system/sigsuspend/sigwait*/pause cancellation-point Zig re-ports.
• pthread_create/join/key_create/atfork/cleanup_push/getattr_np and arch clone/tls/unmap/thread-area pieces stay musl C/asm.
• Futex mutex/cond/rwlock cluster stays musl C: pthread_mutex_{lock,timedlock,trylock,unlock}, pthread_cond_{broadcast,destroy,init,signal,timedwait,wait}, __lock.c, lock_ptc.c, __wait/vmlock. This cluster previously broke multithreaded stage4 with mutex/condition_variable aborts, so it was intentionally not touched in this safe increment.

VERIFICATION:

• python3 scripts/check_musl_migration.py: OK.
• 21 non-loongarch target compile-checks passed in Debug, ReleaseFast, ReleaseSmall.
• Built stage4 from this branch.
• stage4 cross-asm build-exe C smoke passed for all touched setjmp families/variants.
• stage4 threaded x86_64 test-libc-nsz -j8 repeated 3x: success, no mutex/thread aborts.
• Full test-libc-nsz -Dtest-filter=all passed for x86_64 plus qemu spot checks: arm, thumb, riscv, powerpc, s390x.