[llvm] [X86] Disable the `vpdpwssd -> vpmaddwd + vpaddd` combiner pattern on AMD Zen4 (PR #84347)

[Simon Pilgrim via llvm-commits]

https://github.com/RKSimon commented:

I'm hoping we can address this for the general case in #84182 - but in the meantime, we shouldn't be using CPU name matching - its better to create a "TuningFastVNNI" flag in X86.td and add it to the ZN4Tuning flags and use that to control the MC pattern.

But hopefully this patch won't be necessary..........

https://github.com/llvm/llvm-project/pull/84347