[llvm] LoopVectorize: fix phi cost when it is scalar after vectorization (PR #74456)

[Ramkumar Ramachandra via llvm-commits]

artagnon wrote:

Fixed commit message.

> It may be good to also add a use of the recurrence to the test, the load and GEP will get DCE'd.

I don't think this would apply, since nothing gets vectorized in the minimal X86 example, and I don't think we need an example showing vectorization to fix the bug.

https://github.com/llvm/llvm-project/pull/74456