docs(wiki): phase B training path and hardware decisions — recorded, not built

Claude-Session: https://claude.ai/code/session_01FWncR69HgGPuei1dLrW3cU

wiki/concepts/vision-review-outbox.md

@@ -106,7 +106,8 @@ Three surfaces, nothing else — it must not grow into a fleet console:
 - **`GET /export/labels.csv`** — reviewed, usable rows: item, booth, crop path, the reviewer's
   label, the operator's category + classes, the camera's class + confidence, downgraded, at.
   Crops are not packaged: the phase-B trainer runs **on the same host** (its GPU) and reads them
-  off the volume — `docker-compose.collector.yml` carries the `trainer` seam as a commented
+  off the volume ([[bodytype-classifier-training]]: CPU-only, the Xeon is enough) —
+  `docker-compose.collector.yml` carries the `trainer` seam as a commented
   `profiles: [train]` one-off job (next increment).
 
 **Deploy notes.** Bind the published port to the host's **Netbird address** (`COLLECTOR_BIND`),