[turbofan] Introduce prediction for exception handlers.

test/unittests/compiler/loop-peeling-unittest.cc

Index: test/unittests/compiler/loop-peeling-unittest.cc
diff --git a/test/unittests/compiler/loop-peeling-unittest.cc b/test/unittests/compiler/loop-peeling-unittest.cc
index d3eff716a7f0aabc3c2b3197995b2593e23dfce5..c725a27cc0e9c7271633c9ac54bbc0e25fddcc64 100644
--- a/test/unittests/compiler/loop-peeling-unittest.cc
+++ b/test/unittests/compiler/loop-peeling-unittest.cc
@@ -478,7 +478,8 @@ TEST_F(LoopPeelingTest, TwoExitLoopWithCall_nope) {
 
   Node* call = graph()->NewNode(&kMockCall, b1.if_true);
   Node* if_success = graph()->NewNode(common()->IfSuccess(), call);
-  Node* if_exception = graph()->NewNode(common()->IfException(), call);
+  Node* if_exception = graph()->NewNode(
+      common()->IfException(IfExceptionHint::kLocallyUncaught), call);
 
   loop->ReplaceInput(1, if_success);
   Node* merge = graph()->NewNode(common()->Merge(2), b1.if_false, if_exception);