[llvm] [LV] Explicitly disable in-loop reductions for AnyOf and FindIV. nfc (PR #163541)

[Mel Chen via llvm-commits]

================
@@ -1248,6 +1243,13 @@ unsigned RecurrenceDescriptor::getOpcode(RecurKind Kind) {
   case RecurKind::FMaximumNum:
   case RecurKind::FMinimumNum:
     return Instruction::FCmp;
+  case RecurKind::AnyOf:
----------------
Mel-Chen wrote:

I think we should maintain consistency in how we handle AnyOf and FindIV. In particular, AnyOf, whether it currently only supports integer types or will support floating-point types in the future, should consistently use Instruction::Or in ComputeAnyOfResult to fold each part.

https://github.com/llvm/llvm-project/pull/163541