Handle project suspension in the compute service (non-destructive pause/resume)

[scotwells]

Summary

The platform is introducing project suspension — a reversible, non-destructive control that pauses everything a project is running and resumes it intact on reinstatement (platform enhancement: datum-cloud/enhancements#800).

Compute is the launch-blocking integrator. This issue covers all compute-side work to honor project suspension losslessly.

Acceptance criteria

• Compute enhancement/design doc — pause/resume behavior, reinstatement semantics + expected resume latency, and interaction with the instance pause primitive.
• Instance snapshot/suspend/resume primitive — lossless pause of a single instance (in-memory + disk state).
• Honor project suspension via the service-catalog Suspend/Resume hooks — pause instances and stop serving on suspend; retain disks, config, and allocations; resume from preserved state on reinstate; never take the destructive teardown path used for service disable.
• Report per-consumer paused/resumed status back to the platform rollup.
• Emit standardized control-plane Events (ProjectPaused / ProjectResumed / PauseFailed) and ship compute's activity rendering for them.
• Graceful PauseFailed degradation if the pause primitive is unavailable — degrade to least-destructive, never silently delete.
• Conformance test — suspend → reinstate → verify instance state preserved.

Consumer-facing documentation for compute suspension is tracked in datum-cloud/docs (with input from this issue), not here.

Dependencies

• Platform primitive (ProjectSuspension + propagation) — milo-os/milo
• Suspend/Resume integration contract — milo-os/service-catalog

---

Part of the Project Suspension & Lifecycle Controls enhancement (datum-cloud/enhancements#800). Rendered proposal: https://github.com/datum-cloud/enhancements/blob/enhancement/project-suspension/enhancements/platform/project-suspension/README.md