[llvm] [SLP]Improve findReusedOrderedScalars and graph rotation. (PR #77529)

[via llvm-commits]

eaeltsin wrote:

[reduced3.zip](https://github.com/llvm/llvm-project/files/14450664/reduced3.zip)

please see the reproducer attached. 

the difference I see after compiling with `-O3` is:
```
   %5 = select <4 x i1> %2, <4 x float> %3, <4 x float> %4
-  %6 = shufflevector <4 x float> %5, <4 x float> poison, <4 x i32> <i32 2, i32 0, i32 3, i32 1>
+  %6 = shufflevector <4 x float> %5, <4 x float> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 0>
   store <4 x float> %6, ptr %tmp.i.i4, align 8
```


https://github.com/llvm/llvm-project/pull/77529