[llvm] [SLP] Emit reduction instead of 2 extracts + scalar op, when vectorizing operands (PR #147583)

[Alexey Bataev via llvm-commits]

alexey-bataev wrote:

> @alexey-bataev I see that you relanded this change, but don't see any description here or in the commit as to what changed. Can you briefly describe?

Prevented the emission for the case, when it tried to emit reduction for non-immediate operands.
I.e. if you have something like `(a+b) + (c+d)` and the vectorizer prefers to vectorize pair `(a,c)`, it should not be emitted as a reduction, still need to emit as extractelements

https://github.com/llvm/llvm-project/pull/147583