Sub-process Approach for Agent

[michaelquigley]

This is a sub-task of #463.

In working through the agent implementation I figured out that some of the Caddy-based backends really don't want to allow running multiple instances per-process. The caddy.Run() call really expects to only happen a single time per-process.

Also, it might be a little less than ideal to lose all of the running shares in a case where an agent process crashes.

So, let's explore a sub-process based agent approach. Instead of having a single large process to manage all of the agent frontends and backends, let's make the core agent into a sub-process manager.

Spawning processes is relatively easy. There may be challenges around discovering which processes correspond to existing shares, and detecting crashed sub-processes.

[michaelquigley]

The os.Process.Kill() method sends SIGKILL, which cannot be properly caught with a shutdown hook. Need to either post-cleanup in the daemon, or explore sending SIGQUIT and waiting to join.