fix(PrimaryClient): clear sticky stop flag before deliberate reconnect

reconnectStream() did stream_.close() + stream_.connect() without clearing
the sticky stop_requested_ cancellation flag introduced in
5b22c97 ("make connection attempts cancellable to unblock teardown").

PrimaryClient::stop()/~PrimaryClient() set the flag via stream_.requestStop()
so a producer thread stuck in TCPSocket::setup() aborts promptly at teardown.
The flag is sticky and only cleared on (re)start (URProducer::setupProducer,
RTDEClient::init). A deliberate reconnect via reconnectStream() never cleared
it, so after stopPrimaryClientCommunication() any resendRobotProgram() call
failed: TCPSocket::setup() bailed out at the up-front stop_requested_ check and
logged "Failed to reconnect primary stream!".

This regressed UrDriverTest.send_robot_program_retry_on_failure on every
integration runner. Clear the flag in reconnectStream() before connecting,
mirroring URProducer::setupProducer(); a deliberate reconnect is a restart and
must not honor a stale teardown cancellation.

Co-authored-by: Rune Søe-Knudsen <urrsk@users.noreply.github.com>

Author: urrsk

src/primary/primary_client.cpp

@@ -551,6 +551,10 @@ bool PrimaryClient::reconnectStream()
 {
   URCL_LOG_DEBUG("Closing primary stream...");
   stream_.close();
+  // A deliberate reconnect must clear any sticky cancellation left by a prior stop()/teardown
+  // (stream_.requestStop()), otherwise TCPSocket::setup() aborts immediately and the reconnect
+  // fails. Mirrors URProducer::setupProducer() which clears the flag before (re)connecting.
+  stream_.clearStop();
   if (stream_.connect())
   {
     URCL_LOG_DEBUG("Primary stream connected");