Do not wait on rx_task and tx_task when closing the socket.

[Hugal31]

When loosing the connection to a peer, if the send buffer happens to fill up before the keep alive timeout, the tx_task can be stuck in the send_batch operation. On my computer, it can be stuck for 15 minutes before returning with EHOSTUNREACH ("No route to host").

Now this is not that bat itself, it's just a dangling task, TransportLinkUnicastUniversal and TransportUnicastUniversal. But this make the race condition mentionned in #1886 very likely: if the peer reconnect within those 15 minutes, the reconnection is bogus.

This "fix" make TransportLinkUnicastUniversal::close() not wait for the rx and tx task before returning. It doesn't seem to cause any issue. I didn't do the same in TransportUnicastLowlatency because I am not sure it is necessary.

Now I am not really happy with this hack, but I would like to have a solution for #1886 and it's a bit too involved for me alone.

An alternative would be to watch for the cancellation token while calling send_batch.

[github-actions]

PR missing one of the required labels: {'dependencies', 'internal', 'new feature', 'enhancement', 'documentation', 'bug', 'breaking-change'}

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 71.20%. Comparing base (1a10597) to head (2a2f45e).
Report is 12 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1896      +/-   ##
==========================================
+ Coverage   71.18%   71.20%   +0.01%     
==========================================
  Files         364      364              
  Lines       65647    65647              
==========================================
+ Hits        46733    46741       +8     
+ Misses      18914    18906       -8     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.