Inaccurate after_unwind hook in C2 exception handler

JDK-8265702 changed the `process_frames` argument in the construction of the `RegisterMap` in `OptoRuntime::handle_exception_C()` from `true` to `false`. That should be ok, since we don't touch any oops in that context. However, that changed caused intermittent failures in testing. JDK-8265984 later reverted that change and worked around it by instead adding a ThreadWXEnable scope in `ZBarrierSetNMethod::nmethod_entry_barrier()`. However, it seems like the original fix in JDK-8265702 might have uncovered some other issue that we should investigate further as it suggests that we might not be processing a frame somewhere else where we should.