aarch64: Use memcpy_simd as the default memcpy

Since __memcpy_simd is the fastest memcpy on almost all cores, replace
the generic memcpy with it.

(cherry picked from commit 91ac82d0)