agentd: remove workspace abstraction, improve list-runs/logs/status CLI

Overview

Remove the outdated --workspace abstraction from agentd CLI and improve the list-runs, logs, and status commands to use the new /var/log/agentd/<run-id>/ structure.

Changes Needed

1. agentd list-runs (no args required)

• Scan /var/log/agentd/ for all run directories
• Display: run-id, start time, completion status, total cost & iterations
• Completion status determined by parsing events.jsonl for terminal events:
• EventComplete → "completed"
• EventError → "error"
• EventGuardrail → "guardrail"
• No terminal event → "running"

2. agentd logs <run-id>

Hybrid approach:

• If container is running: pass through to podman logs
• If container stopped: read /var/log/agentd/<run-id>/events.jsonl and pretty-print events (EventThinking, EventToolCall, etc)

3. agentd status <run-id> (remove --workspace arg)

• Read from /var/log/agentd/<run-id>/events.jsonl
• Show completion status (parse last event)
• Show cost/tokens/iterations
• Show any errors/guardrails hit

4. Remove workspace abstraction

• Remove --workspace flag from all commands
• Update all code that references workspaces
• Use run-id as primary identifier throughout

Rationale

• Simplifies UX: users only need run-id, not workspace paths
• Uses persisted logs for completed runs (container logs may disappear)
• list-runs becomes the discovery mechanism for historical runs
• logs command provides human-readable event replay instead of raw json