gh-148501: Disable AArch64 Mach-O LOH for JIT stencils - #157040

Currently, LLVM’s AArch64 Mach-O linker optimisation hints prevent the JIT optimiser from folding adrp/ldr constant loads into movz. These hints are actually unused because stencils are not directly passed to the Mach-O linker.

This PR disables the linker hints on macOS Aarch64 with -aarch64-enable-collect-loh=false.

E.g. the stencil for _LOAD_SMALL_INT_r01 on macOS currently looks something like this:

Lloh0:
jit_temp_0_x8:
    adrp    x8, __JIT_OPARG_16@GOTPAGE
Lloh1:
    ldr     x8, [x8, __JIT_OPARG_16@GOTPAGEOFF]
    and     x8, x8, #0xffff

Lloh2:
    adrp    x9, __PyRuntime@GOTPAGE
Lloh3:
    ldr     x9, [x9, __PyRuntime@GOTPAGEOFF]

    add     x8, x9, x8, lsl #5
    mov     w9, #13889
    add     x24, x8, x9

    .loh AdrpLdrGot Lloh2, Lloh3
    .loh AdrpLdrGot Lloh0, Lloh1

After this PR, the linker optimisation hints are removed and the JIT optimiser can correctly fold adrp/ldr constant loads.

    movz    x8, 0

    adrp    x9, __PyRuntime@GOTPAGE
    ldr     x9, [x9, __PyRuntime@GOTPAGEOFF]

    add     x8, x9, x8, lsl #5
    mov     w9, #13889
    add     x24, x8, x9

Overall, this reduces the size of all stencils on macOS by 4.5%:

I ran the pyperformance suite before and after on a Apple M5 Pro which suggests a geometric mean improvement of 0.57% and up to 7.3% faster on scimark_sparse_mat_mult.

Benchmarks and code size analysis was run by GPT 5.6 Sol. Code changes are authored by myself.

I realised too late that work on the JIT is currently paused. I'm still opening the PR since #148501 is labeled as a bug, but feel free to ignore it until there's an official decision on the PEP by the steering council.

A follow up to this would be to recognise the @GOTPAGE/@GOTPAGEOFF Darwin syntax similar to what's done for the :got:/:got_l12: ELF syntax so patch_aarch64_33rx can deal with it.

• Issue: Poor quality stencils produced with Homebrew clang version 22.1.2 on Apple Silicon #148501