[clang] [AArch64] Implement FP8 floating-point mode helper intrinsics (PR #100608)

[via cfe-commits]

https://github.com/SpencerAbson commented:

I'm unsure as to whether the fpm format will ever be changed or extended, but there are some things to consider if it is:
- using a 64 to 32-bit downcast to zero-out bits 32-37 might be a difficult bug to find, and the `__scale` parameter here is not masked to keep it in range. 
- passing `__arm_fpm_init()` as the `__fpm` to these CodeGen tests cannot check whether the correct value is used to zero-out the target bits.




https://github.com/llvm/llvm-project/pull/100608