[x86] split more 256-bit stores of concatenated vectors

As suggested in D62498 - collectConcatOps() matches both
concat_vectors and insert_subvector patterns, and we see
more test improvements by using the more general match.

llvm-svn: 362620