List Agents

Return a flat list of agent sessions for the web GUI grid.

When TaskStore.agents is empty (e.g. only mock adapters spawned and they never heartbeat) we fall back to synthesising one entry per claimed/in-progress task, marked with "synthetic": true. That keeps the GUI grid populated during demos and avoids the dreaded "0 sessions" empty state when work is obviously in flight.

Both branches render the id and title of the task a session is on, so both narrow to the caller's tenant scope: the live sessions carry no tenant of their own and are placed by the task they name, while the synthesised entries are built from a scoped read in the first place.