docs: add CHANGELOG.md, sync all guides with config.yaml migration

- CHANGELOG.md (Keep-a-Changelog, dated entries from git history)
- README: production pipeline framing, config table, layout, flags, changelog link
- configuration.md: legacy env overrides, zones geometry-only, archive paths
- scripts.md: current flags, correct archive/ paths, tracked vs ignored weights
- deployment.md: config.yaml in sync list, model_mode location, /api/model-modes
- models.md + architecture.md: config.yaml pointers, deprecated src/config.py

docs/scripts.md

@@ -4,9 +4,10 @@
 - `predict.py` — **single entrypoint: production service (zero flags) + dev CLI.**
   Combined sack+truck model + repo `src/` modules, shapely zones,
   visual-similarity/Re-ID debounce, SQLite, live-frame publish. Run by
-  `karung-counter.service`. Flags: `--source/--env/--model/--output-dir/
-  --output-json/--sack-conf/--truck-conf/--box-conf(placeholder)/
+  `karung-counter.service`. Flags: `--source/--env/--config/--model/--model-mode/
+  --output-dir/--output-json/--sack-conf/--truck-conf/--box-conf/--box-model/
   --batch-timeout/--max-frames/--no-dashboard/--no-db` (see `predict.py --help`).
+  Config: `config.yaml` (canonical) + `.env` (secrets/deployment).
 - `src/main.py` — deprecated shim: prints a pointer to `predict.py`, still runs the
   old v3 loop for backward compat.
 - `archive/` — retired: `predict_new.py` (duo-model state machine),
@@ -14,19 +15,23 @@
   (minimal CPU demo), `ini.py`, check/merge/test scripts. Not imported anywhere.
 
 ## Calibration & data tools
-- `get_coordinates.py` / `get_calib_frame.py` — zone calibration (`calib_frame.jpg`,
-  `zones.json`); `capture_frame_native.py` — grab native frames (`live_frame_native.png`).
-- `extract_random_frames.py` — sample frames for labeling; `check_*.py`
+- `archive/get_coordinates.py` / `archive/get_calib_frame.py` — zone calibration
+  (`calib_frame.jpg`, `zones.json`); `archive/capture_frame_native.py` — grab
+  native frames (`live_frame_native.png`).
+- `extract_random_frames.py` — sample frames for labeling; `archive/check_*.py`
   (`check_20_23`, `check_20_aug`, `check_all_after`) — batch verification passes.
-- `test_bytetrack.py`, `test_track_crash.py` — tracker experiments/crash repro.
+- `archive/test_bytetrack.py`, `archive/test_track_crash.py` — tracker experiments/crash repro.
 
 ## DB / ops helpers
-`counter_dashboard.py` (dashboard), `backup.py`, `dump_db.py`, `check_jetson_db.py`,
-`migrate_jetson_db.py`, `merge_batches_20_23.py`, `merge_batches_28_35.py`,
-`update_batches.py`, `diagnose_truck_jetson.py`, `deploy_to_jetson.py`,
-`export_model.py` / `export_v4.py` (TensorRT export).
+`counter_dashboard.py` (dashboard), `check_jetson_db.py`,
+`archive/backup.py`, `archive/dump_db.py`,
+`archive/migrate_jetson_db.py`, `archive/merge_batches_20_23.py`,
+`archive/merge_batches_28_35.py`, `archive/update_batches.py`,
+`archive/diagnose_truck_jetson.py`, `deploy_to_jetson.py`,
+`export_model.py` / `archive/export_v4.py` (TensorRT export).
 
 ## Ignored at runtime (`.gitignore`)
 `.env`, `*.db`, state JSONs, `batch_history_folder/`, media (`*.mp4/*.jpg/*.png`),
-model binaries (`*.pt/*.onnx/*.engine`), venvs, IDE files — so the large weights,
-videos and local DBs in this working tree are intentionally untracked.
+`*.engine` (rebuildable via `export_model.py`), venvs, IDE files — so the large
+weights, videos and local DBs in this working tree are intentionally untracked.
+(`models/*.pt`/`*.onnx` ARE tracked so clones are inference-ready.)