[FIRRTL] Optimize foldFlow, NFC.

Instead of scanning the results of an instance to find the result
number (which is O(n) in # results) use OpResult to get it directly.

This is pretty hot in the profile because it is invoked frequently
by the verifier.  This change sped the full end to end (chirrtl to
verilog) time on a large testcase by 3%.

lib/Dialect/FIRRTL/FIRRTLOps.cpp

@@ -123,14 +123,10 @@ Flow firrtl::foldFlow(Value val, Flow accumulatedFlow) {
       .Case<RegOp, RegResetOp, WireOp, MemoryPortOp>(
           [](auto) { return Flow::Duplex; })
       .Case<InstanceOp>([&](auto inst) {
-        for (auto arg : llvm::enumerate(inst.getResults()))
-          if (arg.value() == val) {
-            if (inst.getPortDirection(arg.index()) == Direction::Out)
-              return accumulatedFlow;
-            else
-              return swap();
-          }
-        llvm_unreachable("couldn't find result in results");
+        auto resultNo = val.cast<OpResult>().getResultNumber();
+        if (inst.getPortDirection(resultNo) == Direction::Out)
+          return accumulatedFlow;
+        return swap();
       })
       .Case<MemOp>([&](auto op) { return swap(); })
       // Anything else acts like a universal source.