[llvm] [RISCV] Account for VL toggle cost when splitting VP memory intrinsics (PR #223992)

[Luke Lau via llvm-commits]

lukel97 wrote:

> I guess not really a question for LoopVectorizer, but do we have a similar VL toggle problem when splitting fixed vectors?

For VP intrinsics yes, but I think this PR should also handle fixed vectors too since getTypeLegalizationCost should work on both.

For non-VP loads and stores I don't think it would introduce toggles since we're splitting to the same sized vector, the same as the scalable case. 

https://github.com/llvm/llvm-project/pull/223992