[llvm] [RISCV] Increase default vectorizer LMUL to 8 (PR #222213)

[Olaf Bernstein via llvm-commits]

camel-cdr wrote:

I think the correct approach for the future is:

generate a max LMUL without spilling loop, if that is above LMUL=2 also generate a LMUL=2 tail loop, unless the target supports Ovlt.

The max LMUL loop could also ommit tail-folding.

https://github.com/llvm/llvm-project/pull/222213