[llvm] Reland "[RISCV][TTI] Enable masked interleave access for scalable vector (#149981)" (PR #151665)

[Luke Lau via llvm-commits]

https://github.com/lukel97 approved this pull request.

LGTM, I tested it out locally and can confirm its generating masked vlseg/vsseg as expected from the loop vectorizer, both for internal predication and with EVL tail folding.

I think the last two restrictions in the PR description just need updated now that they're removed.

https://github.com/llvm/llvm-project/pull/151665