fix: Clear adapter send queues on client disconnection

[simon-lemay-unity]

Likeliest cause of MTT-2329.

When the server disconnected a client, it would clear the send queues related to that client. But the reverse was not done. When the client disconnected from the server, it would not clear its send queues. If there was data in there, the adapter would attempt to send it on its next update (or if properly shut down, it would try to send it the next time a connection was made using the same client ID).

The fix is simply to apply the same logic when client disconnects as when server disconnects.

Changelog

com.unity.netcode.adapter.utp

• Fixed: Fixed issue where disconnecting from the server with data still in the queue would result in an error message about a stale connection.

Testing and Documentation

• Includes integration tests.
• No documentation changes or additions were necessary.

[NoelStephensUnity]

@simon-lemay-unity Take a look at #1653, I think some develop transport PRs didn't make it into the v1.0.0 branch, but I did have better success with some of the modifications I made up to the point I realized all of those tests are #IF'd out in develop with a todo note about getting them working on consoles. It is the same timing issue come back to haunt you! :)

[simon-lemay-unity]

@simon-lemay-unity Take a look at #1653, I think some develop transport PRs didn't make it into the v1.0.0 branch, but I did have better success with some of the modifications I made up to the point I realized all of those tests are #IF'd out in develop with a todo note about getting them working on consoles. It is the same timing issue come back to haunt you! :)

Yeah, I may have not been as diligent backporting PRs when they only impacted tests. Thanks for bringing the changes in 1.1.0.

Regarding the tests being defined out for consoles, the reason is that at the time we didn't run UTP tests on these platforms. It was deemed a bit useless to investigate adapter test failures when we didn't even know if UTP itself was working correctly on consoles. Now that we know it does, we can get back to enabling these tests and investigating the failures (it's a task in our current sprint).