[X86][SSE] Avoid scalarization of v2i64 vector shifts (REAPPLIED)