chore: separate shutdown and close for ProtocolClient

[kostasrim]

Add ProtocolClient::Shutdown() which does not close the socket.

For cluster:

When a migration is cancelled, we should not Close() the socket inside Migration::Finish(). The reason is that

  195   if (should_cancel_flows) { 
  196     OnAllShards([](auto& migration) {
  197       CHECK(migration != nullptr);
  198       migration->Cancel();
  199     });   
  200     exec_st_.JoinErrorHandler();                                                                   
  201   }                                                                                                
  202 }   

does not block and wait Migration::SyncFb to complete. In this context, error is empty so there is no error handler to join on.

Then within sync fb the dispatched callbacks of OnAllShards execute:

  285     OnAllShards([this](auto& migration) { migration->PrepareFlow(cf_->MyID()); });
  286     if (!exec_st_.IsRunning())
  287       // exit loop here

And initiate IO on a Closed() fd triggering the check fd >0 within the socket destructor.

Replication:

Also similarly, calling sock->close() early can lead to the same crash in dfly flows. The reason is that the soon to be cancelled fiber attempts to do IO and crashes because of the check fail. Now instead, we shutdown, wait for the flows to complete and then close the socket.

resolves #5968 and #5982

[vyavdoshenko]

lgtm