[X86][SSE] Split off shuffle mask canonicalization from lowerVectorShuffle. NFCI.

The new function now returns true if the shuffle should be commuted.

This will allow target shuffle combines to share the code.

llvm-svn: 277728