[llvm] [LV] Support partial reduce subs/fsubs without a mul operand (PR #199664)

[Florian Hahn via llvm-commits]

https://github.com/fhahn commented:

We have the following in TTI, is that still accurate in all cases?

```
   // Integer partial sub-reductions that don't map to a specific instruction,   <<<
  // carry an extra cost for implementing a double negation:
  //      partial_reduce_umls  acc, lhs, rhs
  // <=> -partial_reduce_umla -acc, lhs, rhs
  InstructionCost INegCost = IsSub ? 2 * InputLT.first * TTI::TCC_Basic : 0;
```

https://github.com/llvm/llvm-project/pull/199664