Limits #117900 to only fold when scalar_to_vector doesn't perform implicit truncation, as the scaled shift calculation doesn't currently account for this - this can be addressed in a future update. Fixes #121306
llvm.powi.*
samesign