If you try to build amd64 from an an arm64 machine, you'll get:
ld: error: undefined hidden symbol: blake3_hash_many_neon
and that is because when _ARCH != arm64, the neon file isn't
compiled/linked, but when building the cross tool, it detects
it's on arm64 and automatically enables it.
This copies want is done for the AVX and other x86 specific
optimizations.
With this change, I was successful in building (and booting) an
amd64 world from an arm64 box...