refactor tcp stack and add graceful shutdown

[sdwoodbury]

Overview:

• adds cancellation tokens to the networking stack to gracefully shutdown when iris-mpc terminates and to halt a job when the networking stack experiences an error.
• simplifies the networking stack by removing the peer connection manager and reconnector.
• will make it easier to debug and correct errors related to re-establishing connectivity (connections re-establish but the sessions time out).

This PR initially set out to add graceful shutdown and job cancellation to the networking stack.

• If iris-mpc is shut down, the networking stack should clean up without emitting false positive error logs
• if the networking stack experiences loss of connectivity during a job, terminate the job early

It became clear that while the process of creating a connection needs to retry, once the connections are established, they need not automatically reconnect. Reconnect can be initiated when the sessions are recreated.

This allows the connection manager and reconnector to be removed. it also allows the connection management code to be simplified.

The logic to connect, handshake, and forward messages is unchanged.