[llvm] [VPlan] Simplify WidenGEP::execute (PR #193543)

[Luke Lau via llvm-commits]

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

LGTM. It looks like the loop invariant checks were added a long time ago, so isSingleScalar (f.k.a isUniformAfterVectorization) has probably become more reliable since. 

Tested llvm-test-suite with this on RISC-V and there were no end-to-end diffs

https://github.com/llvm/llvm-project/pull/193543