[llvm] [InstCombine] Support and/or in `getFreelyInvertedImpl` using DeMorgan's Law (PR #85193)

[Craig Topper via llvm-commits]

topperc wrote:

> > We probably need to revert this one before figuring out what to do about #86880
> 
> topperc mentioned in [#86880 (comment)](https://github.com/llvm/llvm-project/issues/86880#issuecomment-2024110702) that it "post-RA machine copy propagation removed a copy that was needed to make the code correct", so it's probably not worthwhile reverting this innocent patch that just happened to unluckily trigger the bug.

I'm debugging MCP now.

https://github.com/llvm/llvm-project/pull/85193