feat(platform): add a compact task run occupancy read

[yannickmonney]

Fleet occupancy checks currently retrieve the entire task description and review context for each worker. Add an optional task_get view, view: "occupancy", that returns compact task identity, the newest agent run, an optional exact requestedRunId, and the selected workflow state. The default full response keeps its existing fields and values.

The compact path uses the same authenticated scope and project grants. Missing or inaccessible requested runs remain unavailable; malformed or failed reads remain unknown. It returns an observation interval, not a lease or permission to start work. Quarantined workflow state remains visible as a hold.

Validation: 143 focused domain/tool tests and 153 existing bridge/dispatch guards passed, plus scoped lint, formatting and manual-register checks. A synthetic large-context fixture fell from 33,995 to 552 UTF-8 bytes (98.4%); this is fixture evidence, not a universal response-size or latency guarantee. The authenticated Postgres integration cases and native read measurements are added to checkAgentTaskReadTools; their hosted execution and all required exact-head checks remain pending.

Independent source review accepted the 11-file patch. No scheduler, capacity policy, workflow wake logic, or Ops adoption changes are included.

[yannickmonney]

The compact task occupancy read and its tests are included in #4565, merged as 0ee5fd9656dc12f08976155d956ee76fe39f4ea3. Closing this superseded component PR; the shared-file changes from later repairs were preserved.