[llvm] [AArch64][SVE] Use loop.dependence.war.mask in vector.memcheck (PR #175943)

[Benjamin Maxwell via llvm-commits]

MacDue wrote:

> The current code does not check that. From the note, it sounds like something that should be checked?

I now track the common aliment, and reject alignments less than the `AccessSize` (and fall back to normal diff checks). 

https://github.com/llvm/llvm-project/pull/175943