Services dashboard: group cantaconmigo containers under one site (fix abandoned false-positives)

[rdemeritt]

Problem

Services dashboard renders cantaconmigo as 3 separate cards (frontend, backend, postgres), all flagged abandoned. They are one Coolify application.

Root cause

api/src/routes/services.ts:130,139 keys groups by coolify.name — a per-container slug (e.g. frontend-v13w39q44mtejeqki7659rca-190520357886). Coolify listApplications() returns the resource UUID (v13w39q44mtejeqki7659rca). Different namespaces → never matches → all 3 marked abandoned, fragmented into separate cards.

Same UUID is on every container as com.docker.compose.project.

Options memos (already written)

• clients/self/projects/hermithost/specs/grouping-architecture-options-2026-04-29.md
• clients/self/projects/hermithost/specs/grouping-implementation-options-2026-04-29.md

Proposed phased work

Phase 1 — hotfix (~8 LOC, services.ts only)

Swap group key from coolify.name → com.docker.compose.project. UUID matches listApplications() directly. Abandoned check works as-is.

Phase 2 — tri-state liveness model

Replace abandoned: boolean with state: 'live' | 'orphan' | 'unmanaged' | 'degraded':

• owner-system unreachable → live (never orphan)
• no owner labels → unmanaged (terminal)
• mixed group → degraded (covers blue/green leftover)

Phase 3 — hierarchy + ownerKind

Add ownerKind: 'coolify' | 'compose' | 'manual', environment, containerCount. 2-level UI: Coolify project → site → containers.

Phase 4 (deferred) — OwnerResolver abstraction

Pluggable resolver per backend (Coolify, compose, swarm, k8s). Defer until second backend (k8s) lands.

Acceptance

• Cantaconmigo renders as 1 card with 3 containers
• No false-positive abandoned flag while Coolify reachable
• Coolify unreachable → groups stay live, never orphan
• New unit tests in api/test/groupBySite.test.ts cover: 3 containers → 1 group, abandoned semantics, missing labels fallback, PR preview isolation, infra exclusion