Clean up the arm64 _rtld_bind stack creation

[andrew]

Summary

We only need to adjust the stack pointer once on entry. Update to
adjust the stack pointer, then use this to find the offset for each
saved register.

While here add a missing space between two words in a comment.

[kib]

BTW, on amd64 we opted eventually to save all GPRs on rtld bind entry, because there are weird languages that use rtld binding but custom calling conventions.

Also, we do not bother saving FPU registers, instead rtld is compiled with -mno-sse.

libexec/rtld-elf/aarch64/rtld_start.S

99

It is slightly more useful to put .cfi_offset right after the instruction which makes the .cfi statement true, since debugger would see a more correct scope for the frame. This means that mov x29, sp needs to be executed right after stp that saved x29.