[llvm] [SLP] Enable full non-power-of-2 vectorization by default (PR #199450)

[via llvm-commits]

Andarwinux wrote:

I believe the problem on x86 is backend can't lower non-pow2 load/store to [masked](https://github.com/llvm/llvm-project/issues/205006) load/store. Although SLP can do this, cost models wrongly show that's [not](https://github.com/llvm/llvm-project/issues/203756#issuecomment-5272687037) profitable.

https://github.com/llvm/llvm-project/pull/199450