Fix send_exit behaviour

[Debilski]

Fixes a bug (non-existent instance variable) introduced in #890 that caused longer timeouts and potential hangs during exit (including when __del__ is called on a game state).

[Debilski]

(Experimenting a little to find more race conditions; will take care of merging myself later)

[Debilski]

This ended up becoming a tiny bit more involved. Previously, run_game would send an exit message (perhaps including a final game state – which is used by pelita server; or could be used by advanced bots for statistics) to the remote teams (exit_remote_teams) and then RemoteTeam.__del__ would be responsible for cleaning up any potential subprocesses.

The cleanup is now done a little more explicitly with an additional function cleanup_remote_teams. We now have a defined waiting period RemoteTeam.shutdown_timeout that we give a subprocess to shutdown cleanly before we terminate it. (A fallback is still present in RemoteTeam.__del__.)

In between I also experimented with having the remote team return a confirmation message after the exit request but it turned out to be a little useless for now.

Developer notes: The tricky part with debugging code like this is to always have an eye on the runtime of pytest. We typically want the tests to run as speedy as possible but there are some tests in which we actually expect a timeout to trigger (without this being explicitly mentioned in the test).

[Debilski]

Merging. Still not optimal (we could take more advantage of the state machine in the network code) but I’m somewhat confident that this leads to fewer random CI failures.