[llvm] clastb representation in existing IR, and AArch64 codegen (PR #112738)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> I was following a pattern of existing TLI functions (starting with `shouldExpandGetActiveLaneMask` 3 years ago). If that's the wrong way to do it, then I guess we have some cleanup to do later in addition to removing the function from this patch.
> 

It's developing an ad-hoc, parallel legalization mechanism to the regular one that happens in a different place. This expands the surface area of developing a target, and I have endless trouble fighting with non-trivial logic in the builder. All the legalization decisions should be made during legalization, not before 

https://github.com/llvm/llvm-project/pull/112738