Finalization issues with latest libp2p master

[dryajov]

This issues is here to track the ongoing efforts to fix finalization.

• Commit vacp2p/nim-libp2p@a52763c broke NBC finalization.
• The specific change that seems to have introduce this behavior was the removal of the delayed send functionality - vacp2p/nim-libp2p@a52763c#diff-e0b26345ee1b1e15968d513e9b2aad60L125-L156, this is further proved by [DONT MERGE] finalization troubleshooting #1382, which re-adds the "send queue".
• PR libp2p lifecycle hooks and pubsub monitor #1360 attempts to reintroduce delayed sending but this time with a proper timeout (pubsub timeouts tuning  vacp2p/nim-libp2p#295).
• There might be other unrelated issues that are causing finalization to fail, such as idle timeouts introduced to prevent inactive channels from piling up. This also affects pubsub publishing, since I suspect that some peers will not notify of a reset in certain situations (when shutting down for example, resetting is optimistic and there is no way for the sender to know if the channel is still alive, hence I suspect that most implementations will "send and forget" resets), this allows dead channels to further accumulate, hence, timeouts are necessary in the absence of a more sophisticated keepalive/ping mechanism, which is missing from mplex.

[stefantalpalaru]

Tangential, but this is an example of commit squashing making bisection harder.

[arnetheduck]

#1413 addresses this, to a certain extent