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.  If SVE is available, a SVE memcpy will be
used by default (including for Neoverse N2).