[llvm] [RISCV] enable VTYPE before whole RVVReg move (PR #117866)

[Philip Reames via llvm-commits]

preames wrote:

> So what should we measure performance on?

I mean, anything is better than nothing?  The claim that this doesn't effect most workloads seems reasonable, but I'd like to see performance data that actually shows that.  Once we have an established baseline, we can refine to workloads of concern.  

https://github.com/llvm/llvm-project/pull/117866