This avoids using a pair of shufps instructions as the previous lowering
was doing.  Instead, we use movss to copy the element to be inserted
into the lower 32 bits of the destination.

Define InstX8632Movss as a Binop, the class to which it properly
belongs.

BUG=none
R=jvoung@chromium.org, stichnot@chromium.org

Review URL: https://codereview.chromium.org/412353005