build(infra): run ClamAV in compose and on the cluster (closes #191) (#193)

Runs a ClamAV daemon (clamav/clamav:1.4.6) in both compose stacks and the Helm chart, health-gated, with a verify-clamav check (EICAR found, clean OK) in verify-stack. ADR-0036 records the scan-in-domain, fail-closed decision (#190).

closes #191

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

infra/docker-compose.yml

@@ -785,6 +785,26 @@ services:
         condition: service_completed_successfully
     networks: [cg]
 
+  # ClamAV daemon (S-28, ADR-0036): the domain scans uploaded diplomas over clamd's INSTREAM
+  # protocol on :3310 before they reach OpenZaak (S-29). The first start downloads ~300 MB of
+  # signatures with freshclam; the volume keeps them across restarts. clamd holds them in memory
+  # (~1 GB), and a reload would briefly hold two copies — ConcurrentDatabaseReload off prevents
+  # that, at the cost of clamd pausing scans during a signature reload.
+  clamav:
+    image: docker.io/clamav/clamav:1.4.6
+    environment:
+      CLAMD_CONF_ConcurrentDatabaseReload: "no"
+    # The image's own healthcheck (clamdcheck.sh: PING → PONG) polls every 30s; poll faster so
+    # wait-healthy sees it as soon as the signatures are loaded.
+    healthcheck:
+      test: ["CMD-SHELL", "clamdcheck.sh"]
+      interval: 5s
+      start_period: 360s
+    mem_limit: 2g
+    volumes:
+      - clamav-db:/var/lib/clamav
+    networks: [cg]
+
   # ── Observability backplane (S-16a, ADR-0023) ──────────────────────────────
   # Grafana-native stack: Tempo ingests OTLP traces (the .NET services export
   # straight to it — no collector hop, S-16b), Prometheus scrapes service
@@ -836,6 +856,7 @@ volumes:
   projection-db:
   objecttypen-db:
   objecten-db:
+  clamav-db:
   # Config volumes — created and populated out-of-band by infra/seed-config.sh
   # (docker cp), because bind mounts don't reach sibling containers on the CI
   # runner. `external` keeps the names deterministic; the seed step manages them.