Extract standalone pyro_predictor package

[MateoLostanlen]

The detection core (YOLO inference + temporal sliding-window state) was tightly coupled inside pyroengine, which depends on pyroclient, requests, and other
API-related libraries. This made it impossible to use the predictor in external code (evaluation scripts, custom pipelines, CI benchmarks) without pulling
in the full engine stack.

Changes

New package: pyro-predictor/

A self-contained Python package with minimal dependencies (numpy, pillow, onnxruntime, ncnn, opencv, tqdm — no pyroclient, no requests):

• pyro_predictor/predictor.py — Predictor class: YOLO inference + per-camera sliding-window state management (_update_states)
• pyro_predictor/vision.py — Classifier class: NCNN/ONNX model loading and inference
• pyro_predictor/utils.py — box_iou, nms, letterbox and helpers

Both Predictor and Classifier accept a verbose=False parameter to suppress all log output (the package uses a named logger pyro_predictor.* instead of
configuring the root logger, following library best practice).

[codecov]

Codecov Report

❌ Patch coverage is 63.63636% with 16 lines in your changes missing coverage. Please review.
✅ Project coverage is 72.59%. Comparing base (d0a36cd) to head (9b96e87).

Files with missing lines	Patch %	Lines
pyroengine/__init__.py	21.42%	11 Missing ⚠️
pyroengine/engine.py	83.33%	4 Missing ⚠️
pyroengine/utils.py	0.00%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #331      +/-   ##
===========================================
- Coverage    80.19%   72.59%   -7.61%     
===========================================
  Files            7        7              
  Lines          828      602     -226     
===========================================
- Hits           664      437     -227     
- Misses         164      165       +1     

Flag	Coverage Δ
unittests	72.59% <63.63%> (-7.61%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.