Crash avoidance

[lukebarnard1]

Nothing should ever crash the entire nomad process otherwise all current state is lost.

Recovering a crashed session would be impossible because child processes die when the parent crashes.

I think the only option is to catch any error within a subterminal and try to revert to the previously known state, or continue in the current state. The user should probably not be told about these, but they can be logged at least.