AArch64: Optimize strcpy

Unroll the main loop.  Large strings are around 20% faster on modern CPUs.

Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 349e48c0)