feat: per-Backend timeout server + Alerts-Deeplink + Retention + Cert-Prune — v1.3.4

- feat(backends): per-Backend `server_timeout_seconds` (nullable, Default 60s).
  Rendert `timeout server <N>s` im HAProxy-Backend-Block — für langsam
  antwortende Upstreams (KI-/Inferenz-Server mit gepufferter Antwort).
  Migration 0044 (+CHECK 1..86400), Model/Repo/Template/UI + Render-Test.
- fix(ui): Dashboard-Alert-Karte verlinkt auf /alerts?tab=events; Alerts-Seite
  respektiert ?tab= (Deeplink landete bisher auf leerem Channels-Tab).
- feat(scheduler): alert_events-Retention (90d) im täglichen Cleanup-Tick —
  Schutz vor unbounded growth der node-lokalen Health-Event-History.
- fix(cluster): Cert-Sync prunt jetzt lokale .pem die der Primary nicht mehr
  hat (Waisen gelöschter Domains); schützt _default.pem + eigenen Node-Cert.
- fix(security): x/text v0.37→v0.39 (GO-2026-5970, Infinite-Loop; via ACME+goose
  aktiv aufgerufen — govulncheck-Release-Gate).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

management-ui/src/i18n/locales/en/common.json

@@ -592,6 +592,8 @@
     "websocketHint": "On: allow long-lived WebSocket / long-poll connections (Proxmox console, SSH-over-WS, AsyncAPI) — tunnel idle 1h instead of 60s. Off: strict HTTP timeouts.",
     "forceHttp1": "Force HTTP/1.1",
     "forceHttp1Hint": "Disables HTTP/2 (h2) on the backend connection — HAProxy negotiates HTTP/1.1 only. Required for backends that don't support h2 (e.g. older nginx configs without the h2 module, legacy apps).",
+    "serverTimeout": "Response timeout (timeout server)",
+    "serverTimeoutHint": "How long HAProxy waits for this backend's response before aborting. Empty = default (60s). Raise it for slow upstreams that do NOT stream (e.g. AI/inference servers with a buffered response). Note: this is an inactivity timeout — streaming backends (SSE/chunked) usually don't need it.",
     "servers": "Servers",
     "noServers": "no server",
     "noServersWarning": "One or more active backends have no servers configured.",