Update rpc testing to include USE_TENSORPIPE directive (pytorch#68080)

Summary:
Pull Request resolved: pytorch#68080

Fixes pytorch#68002

After FaultyProcessGroupAgent was replaced with FaultyTensorpipeAgent there is now a dependency on Tensorpipe for rpc testing. However, if user does not have USE_TENSORPIPE enabled they will hit an issue such `undeclared identifier 'FaultyTensorPipeRpcBackendOptions'`. This is for testing the faulty agent method so it should not block compilation. Update to wrap the Tensorpipe specific code in a directive.

cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse SciPioneer H-Huang

Test Plan: Imported from OSS

Reviewed By: mrshenli

Differential Revision: D32292861

Pulled By: H-Huang

fbshipit-source-id: 4ffb879860ced897674728200a1831f18fea0a4a

Author: H-Huang

torch/csrc/distributed/rpc/testing/init.cpp

@@ -34,6 +34,7 @@ PyObject* faulty_agent_init(PyObject* _unused, PyObject* noargs) {
   // Import the rpc_module so we can subclass TensorPipeAgent
   py::module rpc_module = py::module::import("torch.distributed.rpc");
 
+#ifdef USE_TENSORPIPE
   shared_ptr_class_<FaultyTensorPipeRpcBackendOptions>(
       module,
       "FaultyTensorPipeRpcBackendOptions",
@@ -125,6 +126,7 @@ PyObject* faulty_agent_init(PyObject* _unused, PyObject* noargs) {
           (std::vector<WorkerInfo>(TensorPipeAgent::*)() const) &
               TensorPipeAgent::getWorkerInfos,
           py::call_guard<py::gil_scoped_release>());
+#endif // USE_TENSORPIPE
 
   Py_RETURN_TRUE;
 }