[InstCombine] missed reducing/canonicalizing add overflow patterns

[rotateright]

Forking this off from the multiplication examples in issue #59217 -
These are all equivalent, but we don't transform them:

define i64 @src(i32 %a, i32 %b) {
  %a8 = zext i32 %a to i64
  %b8 = zext i32 %b to i64
  %m8 = add nuw nsw i64 %b8, %a8
  %hi64 = lshr i64 %m8, 32
  ret i64 %hi64
}

define i64 @tgt(i32 %a, i32 %b) {
  %nota = xor i32 %a, -1
  %ov = icmp ult i32 %nota, %b
  %r = zext i1 %ov to i64
  ret i64 %r
}

declare {i32, i1} @llvm.uadd.with.overflow.i32(i32, i32)

define i64 @tgt2(i32 %a, i32 %b) {
  %s = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
  %ov = extractvalue {i32, i1} %s, 1
  %r = zext i1 %ov to i64
  ret i64 %r
}

https://alive2.llvm.org/ce/z/DX2YCM

[rotateright]

There was a patch underway when I filed this:
https://reviews.llvm.org/D138814
...and looks like it came up before:
https://reviews.llvm.org/D107552