aarch64: optimise unaligned array copy long

The code generate for copy array of long is not optimised for the case where the src and/or dest are not 128 bit aligned.

Some partners HW has improved performance when the address is 128 bit aligned.

This issue proposed optimising array copy long so that all load/store are 128 bit aligned even when the src/dst are not aligned with respect to each other.

A new flag will be provided -XX:+AvoidUnalignedAccesses to enable this. This flag will be turned on by default for partners where it is known that there is a performance benefit.