Switch to ncnn

[MateoLostanlen]

In PR #210, we switched to a less performant model because processing time for a 1024 image was too long.

In this PR, we are returning to the more performant model because we managed to significantly reduce the processing time.

1. We use the NCNN format instead of ONNX, which is more optimized for ARM processors. More information can be found here.
2. We reorganized the core function with asynchronous functions. Now, we only wait for one camera to finish processing while analyzing the predictions from the previous capture.

I couldn't achieve the processing speed of Ultralytics without the library, so I suggest using it for now. However, it is a major dependency, so I will continue working on removing it.

[codecov]

Codecov Report

Attention: Patch coverage is 83.46457% with 21 lines in your changes missing coverage. Please review.

Project coverage is 90.21%. Comparing base (040552c) to head (e457622).

Files	Patch %	Lines
pyroengine/core.py	61.36%	17 Missing ⚠️
pyroengine/vision.py	96.49%	2 Missing ⚠️
pyroengine/engine.py	66.66%	1 Missing ⚠️
pyroengine/sensors.py	95.45%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #217      +/-   ##
===========================================
- Coverage    92.35%   90.21%   -2.14%     
===========================================
  Files            6        6              
  Lines          510      501       -9     
===========================================
- Hits           471      452      -19     
- Misses          39       49      +10     

Flag	Coverage Δ
unittests	90.21% <83.46%> (-2.14%)	⬇️

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.