fix(gateway/discord): REST liveness probe to detect zombie clients (#26656)

[teknium1]

Summary

Discord adapters self-heal after a network outage instead of going silently dark until a manual hermes gateway restart. Salvages #26684 (@xxxigm) onto current main (the adapter moved to plugins/platforms/discord/adapter.py since the PR was opened), with the config moved to config.yaml per the .env-is-secrets policy.

Root cause: when a proxy/NAT wedges the socket without delivering a RST, discord.py's WebSocket reconnect never fires and client.start() spins forever without exiting — so the _bot_task done-callback (which only triggers on task completion) never trips either. The bot stays "offline" in Discord for hours. The done-callback already on main covers the "discord.py gives up and start() exits" case; it cannot see this wedged-but-spinning case.

Changes

• plugins/platforms/discord/adapter.py: out-of-band REST liveness probe. Every discord.liveness_interval_seconds (default 60s) the adapter issues fetch_user(bot_id) — the same REST path as message delivery. After discord.liveness_failure_threshold consecutive failures (default 3) it closes the wedged client and raises a retryable fatal error, which trips the gateway's existing _platform_reconnect_watcher and rebuilds the adapter. Probe is cancelled cleanly on disconnect(). Set either knob to 0 to disable.
• Config lives in config.yaml under discord.*; _apply_yaml_config bridges it to internal env vars the adapter reads, matching the existing HERMES_DISCORD_TEXT_BATCH_* pattern (env-is-secrets-only honored; env still wins for explicit overrides).
• tests/gateway/test_discord_liveness.py (new): probe disabled on interval=0/threshold=0, healthy probe keeps running with no fatal error, threshold failures close client + retryable fatal + handler fired, disconnect() cancels the probe.
• website/docs/reference/environment-variables.md: documents the two internal bridge vars, pointing at config.yaml as the surface.

Validation

	Before	After
Socket wedged behind dead proxy	start() spins forever, bot dark until manual restart	probe detects in ~3×interval → reconnect
Recovery	manual hermes gateway restart	automatic via existing reconnect watcher
Targeted tests	—	tests/gateway/test_discord_liveness.py 5/5 green; existing test_discord_connect.py + test_discord_runtime_failure.py still green
E2E (real imports)	—	config.yaml→env bridge, env precedence, and wedged-socket→retryable-fatal→handler-fired-once all verified

Salvages #26684. Closes #26656.

Infographic

[github-actions]

🔎 Lint report: hermes/hermes-2fe48744 vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 11531 on HEAD, 11521 on base (🆕 +10)

🆕 New issues (6):

Rule	Count
unresolved-attribute	5
unresolved-import	1

First entries

run_agent.py:3014: [unresolved-attribute] unresolved-attribute: Object of type `Self@get_credits_spent_micros` has no attribute `_credits_session_start_micros`
tests/gateway/test_discord_liveness.py:179: [unresolved-attribute] unresolved-attribute: Unresolved attribute `fetch_user` on type `_LiveBot`
tests/run_agent/test_credits_notices_toggle.py:76: [unresolved-attribute] unresolved-attribute: Unresolved attribute `_credits_session_start_micros` on type `AIAgent`
tests/gateway/test_discord_liveness.py:130: [unresolved-attribute] unresolved-attribute: Attribute `fetch_user` is not defined on `None` in union `Unknown | None`
tests/gateway/test_discord_liveness.py:162: [unresolved-attribute] unresolved-attribute: Attribute `is_closed` is not defined on `None` in union `Unknown | None`
tests/gateway/test_discord_liveness.py:17: [unresolved-import] unresolved-import: Cannot resolve imported module `pytest`

✅ Fixed issues (1):

Rule	Count
invalid-assignment	1

First entries

tests/run_agent/test_credits_notices_toggle.py:76: [invalid-assignment] invalid-assignment: Object of type `None` is not assignable to attribute `_credits_session_start_micros` of type `int`

Unchanged: 6066 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.

[alt-glitch]

This was generated by AI during triage.

Related: salvage of #26684 (@xxxigm) onto current main (adapter relocated to plugins/platforms/discord/adapter.py); fixes #26656; part of the gateway-wide zombie-adapter watchdog effort tracked in #32574.