fix(channels): converge on runtime per-channel files + live status in admin

[maksymhryzodub-prog]

Problem

Second half of the Telegram restart-recovery fix (runtime side: CleanSlice/runtime#7). The API read/wrote the legacy data/channels.json while the runtime moved to per-channel data/channels/<type>.json:

• a bot configured from agent chat was invisible in the admin Channels tab and to deploy-time env injection;
• a panel-then-chat history showed — and re-injected on redeploy — the stale panel token;
• the tab showed config presence, not live state, so a dead channel looked healthy.

Full root cause and decisions: specs/003-telegram-restart-recovery/ (spec, research, contracts — included in this PR).

Changes

API (agentChannel slice)

• GET /agents/:id/channels: reads data/channels/telegram.json, falls back read-only to the frozen legacy file for pre-convergence agents (a groups-only per-channel file still defers to legacy — env/panel-configured bots keep working), merges runtime-written data/channels/status.json → connected / statusReason / statusUpdatedAt (null = unknown).
• PUT /agents/:id/channels: writes per-channel files via read-modify-write preserving the runtime-owned groups; omitted channels get {removed: true} tombstones (never deletes — stale pod env vars must not resurrect them). Legacy file is never written again.
• buildAgentEnv: tombstoned channel → no TELEGRAM_* env injection (new spec).

Admin

• Channels tab: live Connected / Disconnected (+reason) / Unknown badge; storage description corrected (was still data/channels.json + wrong restart semantics).

SDK: regenerated for admin and app from the updated swagger.

Version: 0.3.30 → 0.3.31 — release-platform tags v0.3.31 on merge so this reaches prod.

Testing

• api Jest: 132 pass (was 118) — new: per-channel read, legacy fallback, tombstone-beats-legacy, groups-only-defers, status merge (connected: null without status), PUT preserve-groups/tombstone/no-op, env-injection exclusion.
• tsc --noEmit clean.
• E2E on staging after both merges: specs/003-telegram-restart-recovery/quickstart.md scenarios 1–5.

Rollout

Merge after CleanSlice/runtime#7 — agents run ghcr.io/cleanslice/runtime:latest with imagePullPolicy: Always, so the runtime half is live for an agent as soon as it restarts on the new image; already-broken agents heal on their next restart with zero manual steps.

🤖 Generated with Claude Code