feat(grobid): run GROBID on-demand (restart "no") to free ~3.5GB when idle

GROBID is only needed during reference extraction (ra_grobid_backfill.py),
not for normal corpus operations (search, embeddings, wiki). Idle it still
holds ~3.5GB on the 8GB Jetson, so keep it stopped by default.

- infra/docker-compose.yml: grobid restart unless-stopped -> "no" (no auto-start
  on boot; db keeps unless-stopped). Start/stop papers-grobid around an R-A run.
- docs/infra/grobid-jetson.md: add "Run on-demand" section with the start/stop
  workflow, and record the refextract head-to-head evaluation on the Jetson that
  led to keeping GROBID: far worse DOI recall on this math corpus (lutz thesis
  101 usable IDs vs refextract 2) and slower per PDF despite a smaller footprint.

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

docs/infra/grobid-jetson.md

@@ -104,7 +104,36 @@ docker compose logs -f grobid        # wait for "Started ... in N seconds" (mode
 
 The `deploy.resources.limits.memory: 4g` cap in the compose file is honored by
 Compose v2 here (no swarm needed). 3 GB suffices for references; 4 GB leaves
-headroom alongside Postgres.
+headroom alongside Postgres. The `restart: "no"` policy means GROBID does **not**
+auto-start on boot — see "Run on-demand" below.
+
+---
+
+## Run on-demand (stop when idle)
+
+GROBID is only needed **during reference extraction** (`scripts/ra_grobid_backfill.py`);
+nothing else in the stack uses it. Idle it still holds ~3.5 GB, so on the 8 GB Jetson
+keep it stopped and start it only for an R-A run:
+
+```bash
+ssh alfred@192.168.178.103 'docker start papers-grobid'        # ~30s model load
+curl -s --retry 30 --retry-delay 2 --retry-connrefused \
+  http://192.168.178.103:8070/api/isalive                      # wait for: true
+
+# ... run R-A from the Mac ...
+PYTHONPATH=. python scripts/ra_grobid_backfill.py --write
+
+ssh alfred@192.168.178.103 'docker stop papers-grobid'         # reclaim ~3.5 GB
+```
+
+First-time deploy uses `docker compose up -d grobid` (above); afterwards the
+container persists, so `docker start/stop papers-grobid` is all you need.
+
+> Why on-demand rather than a lighter tool: refextract was evaluated head-to-head
+> on the Jetson and rejected — far worse DOI recall on this math corpus (lutz
+> thesis: GROBID **101** usable IDs vs refextract **2**), and it was *slower* per
+> PDF despite a smaller footprint. GROBID stays; running it on-demand removes its
+> only real downside (idle RAM).
 
 ---
 

infra/docker-compose.yml

@@ -44,7 +44,11 @@ services:
     init: true            # reap zombie children (GROBID docs recommend --init)
     ulimits:
       core: 0             # disable core dumps (GROBID docs: --ulimit core=0)
-    restart: unless-stopped
+    # On-demand only: GROBID is needed solely during reference extraction
+    # (scripts/ra_grobid_backfill.py), not for normal corpus operations — start it
+    # for an R-A run, stop it after to reclaim ~3.5GB. "no" = no auto-start on boot.
+    # See docs/infra/grobid-jetson.md ("Run on-demand").
+    restart: "no"
     # GROBID is RAM-hungry; cap usage on the constrained Jetson (Orin Nano, 8GB
     # RAM shared with Postgres). 3GB suffices for references; 4g leaves headroom.
     # Honored by `docker compose up` under Compose v2 (no swarm needed).