[llvm] [X86] Reduce znver3/4 LoopMicroOpBufferSize to practical loop unrolling values (PR #91340)

[via llvm-commits]

ganeshgit wrote:

[matrix_vector_mul.zip](https://github.com/llvm/llvm-project/files/15321617/matrix_vector_mul.zip)
With -unroll-runtime, we can see the attached example gaining. So, let us have the metric @RKSimon suggested. 

https://github.com/llvm/llvm-project/pull/91340