stop producer on destruction (UniversalRobots#472)

When a TCP socket silently fails (e.g. network loss, cable pull) — where
no RST or FIN is delivered —
RTDEClient::~RTDEClient() hangs indefinitely.
The background read thread is inside tryGetImpl's inner while(true)
loop. stream_.read() returns EAGAIN every
~1 second due to the SO_RCVTIMEO set in setupProducer, but:
- !running_ is false (the producer was never stopped)
- stream_.getState() == SocketState::Connected is true (the kernel still
considers the socket connected — no
RST was received)
So the loop hits continue and spins indefinitely.
When stopBackgroundRead() is then called in the destructor, it sets
background_read_running_ = false and
calls join(). But background_read_running_ is only checked in the outer
loop of backgroundReadThreadFunc —
the thread is trapped inside tryGetImpl where only running_ (the
URProducer flag) controls exit. The join()
never returns.
                         
The fix is to call prod_->stopProducer() before stopBackgroundRead().
This sets running_ = false, so within
at most one SO_RCVTIMEO interval (~1 second), tryGetImpl sees !running_
and returns, the background thread
exits its outer loop, and the join() completes immediately.
Note that disconnect() already does this correctly (stream_.disconnect()
+ prod_->stopProducer() +
stopBackgroundRead()), but in the destructor stopBackgroundRead() is
called before disconnect(), so the
  producer is never stopped before the join is attempted.

Author: JohnN193

src/rtde/rtde_client.cpp

@@ -91,6 +91,7 @@ RTDEClient::~RTDEClient()
   {
     reconnecting_thread_.join();
   }
+  prod_->stopProducer();
   stopBackgroundRead();
   disconnect();
 }