[llvm] [Analysis] Extend llvm.experimental.cttz.elts to type-based-cost (PR #184578)

[Philipp Rados via llvm-commits]

prados-oc wrote:

@david-arm The `Transforms/LoopVectorize/AArch64/early_exit_costs.ll` test fails because the new cost doesn't model the scalarization anymore but the lowering as can be found in [SelectionDagBuilder](https://github.com/llvm/llvm-project/blob/7f044944e43eab4f5f52664a9022b1cc083ab258/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp#L8304) .

Is this new cost too low or is it more accurate? I don't have much experience with the AArch64 target so I can't tell.

https://github.com/llvm/llvm-project/pull/184578