[llvm] [WIP][TTI] Refactor improveShuffleKindFromMask to match the pattern order of getInstructionCost (PR #151984)

[Simon Pilgrim via llvm-commits]

RKSimon wrote:

I agree I think that's what we're heading towards - I encountered the non-zero costs identity issue as well while triaging some of the regressions in this patch - to better handle length changing shuffles I think I'm going to have to introduce a SK_Identity enum, and possibly SK_Expand/SK_Compress as well. This is all on my current todo list, assuming DAG freeze handling doesn't completely take over :( 

https://github.com/llvm/llvm-project/pull/151984