Triage error handling and propagation up to UI

[daniloc]

The CLI has gotten more complex over time and there are new error conditions that we could probably handle better. Would be great to:

• Review existing error handling (specifically in the agent side of the CLI; the rest will be deprecated in time)
• Catch errors we currently miss (sometimes we hit the rate limit and we don't signal that at all clearly)
• Otherwise make it clear to users how to recover from a bump in the road