Fix Apple Silicon SIGBUS writing native-module literals (JIT W^X)

[dg1sbg]

Problem

On Apple Silicon the LLVM ORC JITLink code/data slab is mapped MAP_JIT and is write-protected (execute mode) per-thread by default; a thread must call pthread_jit_write_protect_np(false) before it may write into that memory. clasp stores Lisp object pointers into each native module's literals vector, which lives in this JIT memory. Several of those stores were not bracketed by a switch to write mode, so on Apple Silicon they fault with SIGBUS (EXC_BAD_ACCESS code=2, KERN_PROTECTION_FAILURE).

This manifests as a crash in loadltv::attr_clasp_module_native while loading freshly compiled native FASLs during the "Compiling Clasp native image" bootstrap step, and at core::core__literals_vset more generally. On x86-64 and Linux the page is genuinely RWX, so the bug is latent there.

Fix

The helpers JITDataReadWriteMaybeExecute() / JITDataReadExecute() already exist for exactly this purpose. This wraps the three native-module literal write sites with them:

• src/core/compiler.cc core__literals_vset — the store into the literals vector.
• src/core/loadltv.cc op_setf_literals (lits[i] = scf) — replacing a bytecode function with its native simple-fun.
• src/core/loadltv.cc attr_clasp_module_native (lits[i] = value loop) — filling a freshly loaded native module's literals; the get_ltv read stays in execute mode and only the store switches to write mode.

Verification

On Apple M5 / macOS 15 / LLVM 22.1.5: before the fix, (require :asdf) against the bytecode image SIGBUS'd at compiler.cc:213; with the first wrap the fault moved to loadltv.cc; with all three, (require :asdf) loads cleanly and the native image (base.nfasl) builds and boots — which previously aborted at the native-image compile step.

🤖 Generated with Claude Code

[Bike]

Everything here appears to duplicate parts of #1768. I do appreciate all these changes but they would be much easier to merge if better organized.