[llvm] Optimize count leading ones if promoted type (PR #99591)

[via llvm-commits]

v01dXYZ wrote:

Thanks for your review.

Done:

* CGP and SD into two separate commits.
* only uses the `m/mi/sd_match` API. Prefer to use `getNOT/buildNoT` instead of manually creating `AllOnes` and `XOR` (this one is not available for VP though in SD).

The remaining task is the test for CGP (especially for vector types).

https://github.com/llvm/llvm-project/pull/99591