Remove reference to client

Co-authored-by: Neelay Shah <[email protected]>

README.md

@@ -486,7 +486,7 @@ batches are executed. Such backends are called *decoupled* backends.
 
 The decoupled backends use one `ResponseFactory` object per request to
 create and send any number of responses for the request. They must send at
-least one final response, even if it is empty and the client opts out of
+least one final response per request (even if it contains no output tensors).
 receiving this final response. For this
 kind of backend, executing a single inference request typically requires
 the following steps: