[llvm] [InstCombine] Tighten use constraint in factorization transforms (PR #102943)

[Kevin McAfee via llvm-commits]

================
@@ -265,3 +265,27 @@ define i6 @sub_shl_same_amount_partial_nuw2(i6 %x, i6 %y, i6 %z) {
   ret i6 %diff
 }
 
+define i8 @add_shl_same_amount_constants(i8 %z) {
----------------
kalxr wrote:

These new tests aren't actually affected by this patch so there will be no diff, I added them since they show a pattern that wasn't tested before that I want to show as being preserved despite the restrictions added by the change. Should I still pre-commit? Or maybe these should be added in a separate NFC PR?

https://github.com/llvm/llvm-project/pull/102943