[ingvarch]

What changes

urga shows what a Nomad agent logs, at a level, following new lines: the agent urga talks to with :agentlogs (alias :monitor), and any client or server with ctrl-l on its list or its screen. Debugging a cluster no longer needs a shell on the machine.

• The screen is the task log screen's text model, so follow, wrap, /, f, n, w and ctrl-s work as on a task log. v steps through the levels (info, debug, trace, error, warn) and opens the stream again at the new one; the title names the level.
• The past comes first where the cluster keeps it: the last MiB of the agent's log file through MonitorExport (Nomad 1.10.4), filtered to the level, then the live lines from Monitor. An older cluster, an agent without log_file or a refused export gives no past and no error. A server named through the cluster is not asked for its past, because Nomad answers that export with a live monitor that never ends.
• ctrl-l on servers shows only when urga talks to a server agent: a client agent ignores the server it is asked about and would answer with its own log.
• A live log that breaks shows its error on the status line, and a past read that breaks partway shows no past, every time. The first fix also makes a closed log stream check its error once before it ends, which task logs share.
• Streams are closed when the screen is left or the level changes. A token without agent:read sees the cluster's refusal.
• README, docs/clients.md and docs/keys.md say this.

Left for later: file streams can still lose their error the same way (older than this branch, #176).

Closes #155

Checklist

• The test for this change was written first and failed
• make check is green
• The title is a conventional commit: feat(ui): ..., fix(nomad): ...