feat(b3): ml-worker becomes optional — embed-only role, decoupled GPU coordination, cpu-embed switch

The ml-worker's ONLY processing role is now the CPU whole-image embed fallback
(tag_and_embed renamed embed_image — Camie tagging was retired #1189 and the
name kept implying otherwise; videos were already handled agent-style: frame
sampling + mean-pool). Detection/cropping/CCIP stay GPU-agent-only, and their
completion is judged per-pipeline: ccip by gpu_job rows, siglip by concept
regions at the current model version — never by image_record.siglip_embedding.
A CPU embed therefore can NEVER close crop work for the agent (regression test
pins this; only the whole-image 'embed' job, the same artifact, is satisfied).

Making removal actually safe (operator will drop the container):
- GPU-queue coordination (enqueue_gpu_backfill, recover_orphaned_gpu_jobs,
  reprocess_gpu_jobs) moved verbatim to tasks/gpu_queue.py on the maintenance
  quick lane — it lived on the 'ml' queue only by module colocation, which made
  the ml-worker a hard dependency of the whole agent pipeline.
- New ml_settings.cpu_embed_enabled (migration 0074, default ON so agent-less
  installs keep working): OFF stops the four import hooks queueing embed work
  nothing will consume and no-ops the manual backfill; switch lives on the
  renamed 'CPU embedding backfill' card.
- NB heads training / auto-apply still run on the ml image (sklearn) — a stack
  that removes the container gives those up too.

Deploy note: in-flight messages under the old task names are dropped by the
new workers; the 60s orphan sweep + hourly backfill re-fire under the new
names immediately.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CDgx8bQS5YrGRK76v8HUnM

tests/test_api_gpu.py

@@ -127,7 +127,7 @@ async def test_backfill_enqueues_then_is_idempotent(db):
     await _img(db, "c" * 64)
     await _img(db, "d" * 64)
     await db.commit()
-    from backend.app.tasks.ml import enqueue_gpu_backfill
+    from backend.app.tasks.gpu_queue import enqueue_gpu_backfill
 
     n = enqueue_gpu_backfill("ccip")     # sync task, own session
     assert n >= 2
@@ -260,3 +260,33 @@ async def test_errors_endpoint_reports_triage_view(client, db):
     assert item["reason_class"] == "truncated_or_corrupt"
     assert item["triage_status"] is None
     assert item["image_url"].startswith("/images/")
+
+
+@pytest.mark.asyncio
+async def test_cpu_embed_never_blocks_gpu_crop_backfills(db):
+    """B3 invariant (operator 2026-07-02): ccip (detect + character) and
+    siglip (concept crops) completion is judged per-pipeline — gpu_job rows and
+    image_region state — never inferred from image_record.siglip_embedding. So
+    an image the CPU fallback already embedded still gets both crop jobs; only
+    the whole-image 'embed' job (the SAME artifact the CPU path produces) is
+    satisfied by it."""
+    from backend.app.models import MLSettings
+    from backend.app.tasks.gpu_queue import enqueue_gpu_backfill
+
+    img = await _img(db, "7" * 64)
+    cur = (await db.execute(
+        select(MLSettings.embedder_model_version).where(MLSettings.id == 1)
+    )).scalar_one()
+    # As if the CPU fallback already embedded it under the current model.
+    img.siglip_embedding = [0.1] * 1152
+    img.siglip_model_version = cur
+    await db.commit()
+
+    assert enqueue_gpu_backfill("ccip") == 1      # crops still open
+    assert enqueue_gpu_backfill("siglip") == 1    # concept crops still open
+    assert enqueue_gpu_backfill("embed") == 0     # same artifact — already done
+
+    tasks = set((await db.execute(
+        select(GpuJob.task).where(GpuJob.image_record_id == img.id)
+    )).scalars().all())
+    assert tasks == {"ccip", "siglip"}