[llvm] [llvm][CodeGen] Update checking method of loop-carried phi in window scheduler (PR #96288)

[Hua Tian via llvm-commits]

huaatian wrote:

During our internal code review, we discovered that we had missed an equivalent loop-carried phi scenario. Previously, we checked the scenario where: "The virtual register defined by the preceding phi is used by the succeeding phi." There is an equivalent scenario where: "The preceding phi uses the virtual register defined by the succeeding phi." This patch improves the check to cover this scenario.
These scenarios were included in the Hexagon test cases, but since the windowing algorithm was not executed in that test case, no issues were encountered. We have now added the corresponding test cases.
![image](https://github.com/llvm/llvm-project/assets/142874007/9a08f4d2-9912-4d7f-a9e2-a5a47efcdc04)
![image](https://github.com/llvm/llvm-project/assets/142874007/a2dda729-a3c2-401e-8dea-5f5a91b6100b)





https://github.com/llvm/llvm-project/pull/96288