[llvm] Aarcht64 add waw sme clustering (PR #212603)

[Pengcheng Wang via llvm-commits]

wangpc-pp wrote:

> @wangpc-pp I did consider adding support for custom SDep to TabelGen macro-fusion, if you have a real user that would be good, as AArch64 still uses C++.

IIRC, Qualcomm's `Xqci` has a fusion for `mv+mv`, in which two instructions have no data dependency (but I don't know if they have a weaker dependency). This can be a potential user of adding `SDep` to TabelGen macro-fusion. cc @lenary 

https://github.com/llvm/llvm-project/pull/212603