[llvm] [Passes] Move LoopInterchange into optimization pipeline (PR #145503)

[Ryotaro Kasuga via llvm-commits]

kasuga-fj wrote:

Here are the compile-time impact by changing the position.

https://llvm-compile-time-tracker.com/compare.php?from=f02992c483d40325ceac5c3520b0b0fca42f8928&to=21856d54baa0b81035fb16c736cbc71cbb4df87a&stat=instructions%3Au

As mentioned in https://github.com/llvm/llvm-project/pull/124911#issuecomment-3024212383, it reduces the compile-time, particularly for ThinLTO. However, since I'm planning to submit some PRs that modify LoopInterchange to further reduce compile time, it might be better to reevaluate later.

> My only curiousity is if this changes compile-times. 

My concern is that, once it is enabled by default, repositioning it may require additional effort. That said, since it's rarely triggered in practice at this moment, compile time appears to be the only aspect with any measurable data. 

https://github.com/llvm/llvm-project/pull/145503