stage-21: swap inference server to psyc-v5 adapter

v5 trained on 598 ex/task (20× v4's 30), with --defang-frac 0.5 over
the new ThreatFox + MalwareBazaar + OTX corpus. Final train_loss 0.3225
vs v4's 0.7397 (56% reduction), 60m20s wall clock on a 3090.

Live eval before swap:
- severity (botnet, ONLINE): v4 high / v5 high — tied, both correct
- ioc_extraction with defanged input (hxxps://, [.], (.), [dot]):
    v4 kept hxxps:// in output (failed canonicalization)
    v5 returned canonical https:// — defang training paid off
- ioc_extraction on real OTX-style prose (never trained on this shape):
    v5 cleanly extracted 2 domains + 1 IP + 1 SHA256 + 1 CVE

Cockpit /api/inference-status confirms the swap:
  {"online":true,"adapter":"/data/adapters/psyc-v5/final"}

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

docker-compose.yml

@@ -51,7 +51,7 @@ services:
   # `--profile gpu`. Uses the psyc-trainer image (built from Dockerfile.train).
   inference:
     image: psyc-trainer
-    command: ["/scripts/serve_model.py", "--adapter", "/data/adapters/psyc-v4/final", "--host", "0.0.0.0", "--port", "8771"]
+    command: ["/scripts/serve_model.py", "--adapter", "/data/adapters/psyc-v5/final", "--host", "0.0.0.0", "--port", "8771"]
     volumes:
       - ./data:/data
       - ./scripts:/scripts