[llvm] [RISCV] Only calculate ordered reduction with FloatingPoint type. NFC (PR #114180)

[Philip Reames via llvm-commits]

preames wrote:

> Without the check querying the `getArithmeticReductionCost()` by `reduce.add` `reduce.and` ... integer reductions without `reassoc` flag will fall into it. And it will return the cost of the floating point ordered reduction.

Which caller is this coming from?  getTypeBasedIntrinsicInstrCost (the one used by CostAnalysis printer pass), appears to pass nullopt for everything except fadd/fmul.  

https://github.com/llvm/llvm-project/pull/114180