JIT fails to allocate mcode memory on M1 for Tarantool

During running JIT-related tests the trace assembling occasionally fails with the following reason in jit.dump output:

failed to allocate mcode memory

This is happens due to none of probed pseudo-random addresses are suited for JIT engine, because all of them are out of LJ_TARGET_JUMPRANGE (+-27Mb). Looks like this happens due to bunch of loaded .dylib libraries (libcurl, libicu, etc.) so all well-suited addresses are already in use and mmap() just returned to big address.

Also, a bunch of problems I faced while trying to resolve the issue:

1. We can't link with the shared luajit library as far as Tarantool uses internal routines from the luajit, so they won't be provided by the shared library.
2. On M1 due to security reasons we can't link the final executed binary with -no_pie option:

ld: warning: -no_pie ignored for arm64

As a result, we can't set -image_base for it too:

ld: warning: Linking with PIE, -image_base will be ignored

3. Looks like there are no analogues of linker scripts for macos (at least I don't find any suited replacements for ld --verbose, -T option), so we can't just add something like the following in the default linker script:

.text.luajit : {
      . = 0x21000000;
      *libluajit.a:*(.text .text.*)
}