[llvm] [AArch64] Combine and and lsl into ubfiz (PR #118974)

[Cullen Rhodes via llvm-commits]

c-rhodes wrote:

> > The codegen changes seem like an overall win to me. Nice! I had a few comments.
> > Just a general question - did you try seeing if the SHL combine could be done in the general DAGCombiner to get the same effect? It seems like a sensible canonicalisation to do, mirroring what InstCombine does at the IR level. I just wondered if you saw regressions in other targets when doing that, or if something else got in the way? Thanks!
> 
> I didn't no, but I can take a look a that when I revisit this after holidays :)

Moving this to DAGCombiner there are ~180 failures across various backends, as well as a regression in `llvm/test/CodeGen/AArch64/swap-compare-operands.ll` since the AArch64 ISD nodes in one of the pattern if statements have to be removed. 

https://github.com/llvm/llvm-project/pull/118974