[AZ-965] NetVLAD-VGG16 backbone checkpoint + YAML/compose wiring

AZ-965 ships the NetVLAD .pt checkpoint that clears the AZ-839
empty-c10_provisioning.backbones SKIP gate. Pipeline-integration
scaffold — encoder is real, NetVLAD tail is honestly labelled as
untrained.

Composition:

* Encoder (26 keys, encoder.0..encoder.28): torchvision
  vgg16(weights=IMAGENET1K_V1) features [:-2], BSD-3-Clause.
  Real ImageNet-pretrained VGG16 conv stack.
* NetVLAD pool + PCA tail (5 keys: pool.conv.{weight,bias},
  pool.centroids, pca.{weight,bias}): random-init via
  torch.manual_seed(0). NOT trained for visual place recognition.

Total: 149,002,112 params (568.4 MiB fp32, sha256=745c6f29...).
Round-trip verified locally: torch.load(weights_only=True) +
load_state_dict(strict=True) succeed; forward(1,3,480,480) emits
{'vlad_descriptor': (1, 4096) fp32} — matches NetVladStrategy
contract per net_vlad.py:247-251.

Two material discoveries documented in the AZ-965 spec:

1. The NetVLAD-VGG16 architecture already lives in repo at
   src/gps_denied_onboard/components/c2_vpr/_net_vlad_architecture.py
   — we instantiate it and save a state_dict, NOT externally source.
2. The PyTorch FP16 runtime expects a .pt state_dict (NOT .onnx).
   BackboneConfig.onnx_path is a misnomer for NetVLAD: per AZ-321
   design + c2_vpr description.md §1, NetVLAD runs on PyTorch FP16
   (NOT TRT). compile_engine is a no-op sha256+path wrap;
   deserialize_engine does torch.load(weights_only=True) +
   load_state_dict(strict=True).

User skipped Option A/B/C/D/E question — judgment call = Option B
(IMAGENET1K_V1 + random tail) per "use judgment, don't block":
* Option A (Nanne translation) was 5-8 SP, above the 5 SP budget.
* Option B is 3 SP, fits the budget, honestly labelled.
* Option C (pure random) was borderline-dishonest per Real Results.

Files:

* scripts/mk_netvlad_checkpoint.py — deterministic generator.
* models/netvlad/netvlad.pt — 568 MiB, via git-lfs (.gitattributes
  extended for models/**/*.pt, *.onnx, *.engine).
* configs/operator_replay.yaml — c2_vpr + c10_provisioning blocks
  populated; the field literally named onnx_path actually points
  at the .pt for NetVLAD per the runtime semantics noted above.
* docker-compose.test.jetson.yml — ./models:/opt/models:ro bind
  mount added to e2e-runner.
* _docs/03_ip_attribution/netvlad.md — provenance, licence, how-to-
  reproduce, honest scope statement ("NOT a real-retrieval
  checkpoint; ESKF divergence under garbage retrievals is the
  expected next gate").
* _docs/02_tasks/todo/AZ-965_netvlad_onnx_backbone_provisioning.md
  — rewritten to reflect the .pt-not-.onnx + Option B discoveries.

Tier-2 verification follows in a separate commit after the harness
run confirms the empty-backbones SKIP gate clears.

Out of scope (filed as follow-ups):

* Real-retrieval NetVLAD weights (Nanne Pittsburgh-30k translation
  or internal team checkpoint) — separate ticket.
* AZ-840 orchestrator PASSing end-to-end (depends on retrieval
  quality + ESKF stability).
* AZ-963 60s smoke ESKF divergence (independent chain).

Co-authored-by: Cursor <cursoragent@cursor.com>

configs/operator_replay.yaml

@@ -17,11 +17,15 @@
 #   * `SATELLITE_PROVIDER_URL` → c11_tile_manager.satellite_provider_url
 #   * `SATELLITE_PROVIDER_API_KEY` → c11_tile_manager.service_api_key
 #
-# AZ-964 (follow-up, not yet filed): the orchestrator test SKIPs at the
-# next gate because `c10_provisioning.backbones` is empty — no NetVLAD /
-# DINOv2 .onnx file ships with this repo. Populating the backbones list
-# here (and provisioning the matching .onnx + verifying it compiles on
-# Tegra) is AZ-964's scope, not AZ-962's.
+# AZ-965 (2026-05-29): `c10_provisioning.backbones` now declares a
+# single NetVLAD-VGG16 entry pointing at `models/netvlad/netvlad.pt`
+# (568 MiB git-lfs blob; see `_docs/03_ip_attribution/netvlad.md` for
+# provenance — VGG16 encoder = torchvision IMAGENET1K_V1 BSD, NetVLAD
+# pool + PCA tail = deterministic-random untrained). Bind-mounted into
+# the e2e-runner at `/opt/models` via docker-compose.test.jetson.yml.
+# AZ-321 design: NetVLAD runs on the PyTorch FP16 runtime (NOT TRT),
+# so the field literally named `onnx_path` here is actually the path
+# to the `.pt` PyTorch state_dict the runtime consumes.
 
 __top__:
   mode: replay
@@ -49,11 +53,22 @@ c7_inference:
   trtexec_timeout_s: 600
   ort_trt_cache_dir: /var/lib/gps-denied/engines/ort_trt_cache
 
+c2_vpr:
+  strategy: net_vlad
+  backbone_weights_path: /opt/models/netvlad/netvlad.pt
+  netvlad_descriptor_dim: 4096
+  warn_top1_threshold: 0.30
+  # faiss_index_path is overlaid at runtime by
+  # tests/e2e/replay/_e2e_orchestrator.py::write_effective_replay_config
+  # to point at <cache_root>/descriptor.index (the C3 fixture's tmp).
+
 c10_provisioning:
   workspace_mb: 4096
-  # backbones intentionally empty — see AZ-964 for the follow-up.
-  # The AZ-839 fixture skip-gate (conftest.py:594-601) fires here
-  # with a clear message until backbone provisioning lands.
+  backbones:
+    - model_name: net_vlad
+      onnx_path: /opt/models/netvlad/netvlad.pt
+      expected_input_shape: [3, 480, 480]
+      input_name: input
 
 c11_tile_manager:
   # satellite_provider_url + service_api_key flow in from env vars