Update ImgExtractedDetection: Soften confidence check

[HonzaCuhel]

This PR softens the confidence check in ImgDetectionExtended class, so that values like 1.0000001192092896 would pass.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 69.69697% with 20 lines in your changes missing coverage. Please review.

Project coverage is 36.54%. Comparing base (6cce15b) to head (dd6461a).
Report is 2 commits behind head on main.

✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
depthai_nodes/utils/logging.py	71.05%	11 Missing ⚠️
depthai_nodes/ml/messages/keypoints.py	60.00%	6 Missing ⚠️
depthai_nodes/ml/messages/img_detections.py	66.66%	2 Missing ⚠️
depthai_nodes/ml/parsers/utils/yolo.py	66.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #166      +/-   ##
==========================================
+ Coverage   35.88%   36.54%   +0.66%     
==========================================
  Files          70       74       +4     
  Lines        4013     4085      +72     
==========================================
+ Hits         1440     1493      +53     
- Misses       2573     2592      +19     

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

[jkbmrz]

What's the use case of conf. values being higher than 1?

[HonzaCuhel]

The reason behind this is, so that when the model is quantized or the values are just badly rounded, the DAI script won't crash. Which is exactly what happened when I tested YOLOv7 tiny inference on a RVC4 device. The parser returned the following confidences scores: [1.0000001192092896, 1.0000001192092896, 1.0000001192092896, 0.6650391221046448], which resulted in the script crash, as most of the values were > 1.

[klemen1999]

Is there anywhere else we have a similar issue that we could handle inside this PR CC: @sokovninn @aljazkonec1

[kkeroo]

Can we rather clip the values to be between 0 and 1?

EDIT: Yeah I see Klemen already said that

[aljazkonec1]

Is there anywhere else we have a similar issue that we could handle inside this PR CC: @sokovninn @aljazkonec1
Yes, Keypoints message also has similar clipping.

[klemen1999]

Yes, Keypoints message also has similar clipping.

Let's also include same checks and clips there inside this PR as well then.