Attempt to fix JIT icache coherency on Arm64

The JIT compiler was experiencing intermittent failures on Arm64/Apple
Silicon due to missing instruction cache invalidation after patching
branch instructions. When update_branch_imm() modified branch targets
in JIT-compiled code, the CPU's icache wasn't being invalidated, causing
it to execute stale cached instructions instead of the newly patched
ones.

This manifested as non-deterministic test failures, particularly in
compute-intensive benchmarks like the pi calculation test.

The fix adds sys_icache_invalidate() after memcpy() in update_branch_imm
to ensure the icache is synchronized with the data cache after code
modification. This is critical on Arm64 cores which have separate L1
instruction and data caches.

Author: jserv

src/jit.c

@@ -611,6 +611,7 @@ static void update_branch_imm(struct jit_state *state,
     pthread_jit_write_protect_np(false);
 #endif
     memcpy(state->buf + offset, &insn, sizeof(uint32_t));
+    sys_icache_invalidate(state->buf + offset, sizeof(uint32_t));
 #if defined(__APPLE__) && defined(__aarch64__)
     pthread_jit_write_protect_np(true);
 #endif
@@ -2229,6 +2230,7 @@ static void resolve_jumps(struct jit_state *state)
 
         uint8_t *offset_ptr = &state->buf[jump.offset_loc];
         memcpy(offset_ptr, &rel, sizeof(uint32_t));
+        sys_icache_invalidate(offset_ptr, sizeof(uint32_t));
 #elif defined(__aarch64__)
         int32_t rel = target_loc - jump.offset_loc;
         update_branch_imm(state, jump.offset_loc, rel);