feat: grouped server cards with progress bars and sparklines

- Agent sends group name (defaults to hostname) and richer metric notes
  (memory/disk include absolute GB values; CPU note includes system uptime)
- push() accepts group param; server stores it as agent_group on deployments
- Upsert key is now (company_id, agent_group, name) so two agents on the
  same company but different servers don't overwrite each other's CPU/Memory
- Add widget config packs: linux-base, web-nginx, database-postgres/redis,
  docker-host, custom-cmd examples

widgets/database-postgres.yml

@@ -0,0 +1,21 @@
+# PostgreSQL — process check + port + optional replication lag custom check.
+
+widgets:
+  - type: process_check
+    name: PostgreSQL
+    process: postgres
+
+  - type: port_check
+    name: PG Port
+    host: localhost
+    port: 5432
+    timeout: 5
+
+  # Optional: check replication lag (requires psql access).
+  # Exit 0 = ok, exit 1 = lag > threshold, exit 2 = cannot connect.
+  # - type: custom_cmd
+  #   name: PG Replication lag
+  #   cmd: |
+  #     LAG=$(psql -U postgres -tAc "SELECT EXTRACT(EPOCH FROM (now() - pg_last_xact_replay_timestamp()))::int" 2>/dev/null)
+  #     [ -z "$LAG" ] && exit 2
+  #     [ "$LAG" -gt 300 ] && exit 1 || exit 0