[llvm] [RegisterPressure] NFC: Clean up RP handling for instructions with overlapping Def/Use (PR #109875)

[Jay Foad via llvm-commits]

jayfoad wrote:

> This cleans things up a bit and _more accurately models subreg lane liveness_

This suggests that it is not NFC? Is there any way of testing this?

https://github.com/llvm/llvm-project/pull/109875