Generates a special fcache_enter_gonative routine for AArch64 (now it matches other architectures) to restore the stolen register and to avoid clobbering x0 on detach.
The new generated code required adding negative displacement support to XINST_CREATE_{load,store} and tweaking some other defines.
Since on Aarch64 we can't jump through memory or write to the PC, for now we assume we're at an ABI boundary and we clobber a caller-saved register.
Augments the api.detach test to use the _with_stats variant, which reproduces the original crash without the fix.
Fixes #4457 (closed)