minor cleanup in websocket code

While reviewing code for PR 209, I noticed the following function aws-net-websocket-registry.adb:

      procedure Watch (WebSocket : Object_Class) is
      begin
         if Is_Registered (WebSocket.Id)
           and then not Watched.Contains (WebSocket.Id)
         then
            Watched.Insert (WebSocket.Id);
            Count := Count + 1;
            Signal_Socket;
         end if;

      exception
         when others =>
            Unregister (WebSocket);
            raise;
      end Watch;

It is slightly un-optimal since it needs to do two lookups when the socket is not watched yet. Perhaps something like the following would be more efficient:

      procedure Watch (WebSocket : Object_Class) is
          Inserted : Boolean;
          Pos : WebSocket_Set.Cursor;
      begin
         if Is_Registered (WebSocket.Id) then
            Watched.Insert (WebSocket.Id, Pos, Inserted);
            if Inserted then
               Count := Count + 1;
               Signal_Socket;
            end if;
        end if;

      exception
         when others =>
            --  ??? Not clear why we would unregister the socket here ? 
            --  also, it is possible that it is still in Watched at this point, and given the first test in this procedure this
            --  might be unexpected behavior.
            Unregister (WebSocket);
            raise;
      end Watch;

Start by reading aws-net-websocket-registry.adb and the Watch procedure's surrounding exception handling. Compare the proposed single-lookup insertion with the current behavior, especially whether Unregister should remain and how Watched is left on failure. Done means the lookup is not duplicated and the exception behavior is consistent.