[X86] lowerShuffleAsVTRUNC - bit shift the offset elements into place instead of shuffle

This helps avoid issues on non-BWI targets which can end up splitting the shuffles to 2 x 256-bit bitshifts of a smaller scalar width