M6 epic: Guest lifecycle and task operations

[Andreas-Froyland]

Tracking issue. Dependency and status checklists only — implementation stays in narrow linked issues.

Outcome

Bounded start/stop/shutdown/reboot operations run as durable, audited Fleet operations with real UPID task progress.

Supersedes

Replaces the "VM boot/stop" and lifecycle: power-state portions of #3. The legacy UPID polling and certificate-fingerprint tests are preserved as requirements.

Depends on

• M6 VM/LXC associations and guest data
• FM-108/FM-109 — durable operations and worker (M1)

Scope checklist

• Start, stop, shutdown, reboot mapped to durable operations — FM-602 (FM-602 — Add Proxmox guest lifecycle and task operations #103, PR FM-602: Proxmox guest lifecycle and task operations #104): four executor kinds, catalog-level proxmox.operate on both surfaces
• UPID task polling with progress, cancellation, and deadlines — FM-602: Fleet-owned UPID parsing, fixed-interval polling against a deadline starting before the mutation, cancellation observed between polls (remote task left running, recorded honestly)
• Deterministic timeout behavior (no wall-clock-sensitive polling; see the legacy waitForTask flake noted in docs/PLAN.md) — FM-602: a sleep between polls, never a clock race
• Partial-node failure and privilege failure surfaced per resource — FM-602: honest task taxonomy (OK/ERROR/unknown), trust-gate and privilege refusals before any network call

Constrained by

• docs/PLAN.md — M6; ADR-0008
• Mutating APIs return operation handles

Non-goals

• Destructive operations (template/clone/snapshot epic)
• Guest OS provisioning; that is Ansible/Nix/profile work, not Fleet's