On ARM and ARM64, each SEH opcode is supposed to map to exactly one
instruction; the operation of writing x29+x30 with a stack offset
with writeback is called ".seh_save_fplr_x".

Using the right opcode would fix inconsistent unwinding if we would
unwind from near the end of the prologue (which is implausible here).

However, clang/LLVM might also get checking for mismatches between
SEH opcodes and the prolog/epilog ranges
(https://reviews.llvm.org/D131394), mainly to guard against
inconsistencies in the unwind info generated by the compiler, but
that would trigger on this handwritten function too.