This adds a fiber-based relay manager and updates the socket abstraction code to use it. From my local and informal testing it propagates events slightly faster. This model though does allow a stream object that needs to relay from more than one source to both use the new #monitor_sock method and do so without creating a thread for each new monitored socket. In the very near future, this will be used by the MsTds::Channel to relay to and from the IO abstraction that handles the TDS framing for SSL necessary to address rapid7/metasploit-framework#18745. I believe we may also be able to use it for services such as the SOCKS proxy to delagate relaying between sockets to the fiber instead of opening a thread per connection. This would offer a performance benefit when many connections are opened and established.

Testing will be the easiest with a pending Metasploit Framework PR which will include the MsTds::Channel object and steps to ensure Meterpreter channels continue to work as necessary. The new close methods for local/remote were necessary to provide hinting to eliminate a deadlock in Meterpreter that would occur when socket was closed from the remote side (the Meterpreter instance sent a close command to Metasploit Framework) and Metasploit would attempt to close it again by sending a frame back to Meterpreter. This isn't necessary because of course Meterpreter already closed it and caused a dead lock where the Meterpreter instance would be unresponsive for about 15 seconds while the request timed out before it'd work just fine again. Now when the local or remote side wants to close the connection, the appropriate methods can be used to only alert the necessary parties on the other side of the relay.

Requires Ruby 3.0+ where non-blocking fibers were added.