[core] Fixed bug: srt_accept failure may make accepted socket leak

[ethouris]

When the socket picked up from the accept queue turns out to have address that cannot be written back to the sockaddr buffer because it's too small, this function returns an error, but the socket remains valid and could leak this way. As this socket can no longer be returned, it must be set broken. Although theoretically this call could be repeated with fixed value, it's unlikely that an application can do a repeated call and inteligently fix the problem in runtime; instead it will rather fail completely and this way the socket will be left leaked.

Draft until: #2681 (prematurely merged)

[maxsharabayko]

Not using m_PeerAddr to avoid locking s->m_ControlLock?

[ethouris]

Things have changed a bit, please merge #2681 first.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 67.73%. Comparing base (09f35c0) to head (d5a7604).
Report is 80 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1884      +/-   ##
==========================================
+ Coverage   67.20%   67.73%   +0.53%     
==========================================
  Files         102      103       +1     
  Lines       20269    20686     +417     
==========================================
+ Hits        13622    14012     +390     
- Misses       6647     6674      +27     

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