Test coverage

[arnetheduck]

A few areas that appeared not to trigger any test failures when refactoring the code - some may be covered partially but might need more variation:

• [ ] Peer disconnection - ensure that cleanup happens in gossipsub, connection manager etc
• [ ] Peer reconnection - ensure that after disconnection and reconnection, peer gets messages again
• [ ] Error handling at each level of protocol negotiation - in particular, simulate failures during:
  • tcp connection (connect to inexistent peer, break off incoming connection without sending data)
  • secure negotiation (no secure protocol, failures while negotiating etc)
  • multiplexer negotiation (invalid channels etc)
  • multistream negotiation (missing handshakes, failed selection etc)
  • failed identify (different peer id etc)
  • breaking protocol limits (too large mplex frame, too large multiselect negotiation message without \n)
• [ ] connection limits - too many connections to same peer
• [ ] mplex close sequence - order of events such as CloseEvent coming after closemessage etc
• [ ] gossipsub write limits - disconnect on too much in channel send log

[sinkingsugar]

Done so far on my side:

• connect to inexistent peer
• secure negotiation (no secure protocol, failures while negotiating etc)

[dryajov]

Related to #370, still needs to be revised.