Add sigmoid smoothing to instance segmentation post-processing of YOLO models family in inference-models

[PawelPeczek-Roboflow]

What does this PR do?

Change to align post-processing for instance-segmentation with the state of old inference - using sigmoid(...) smoothing and 0.5 thresholding.

Type of Change

• Bug fix (non-breaking change that fixes an issue)

Testing

• CI
• I have tested this change locally
• I have added/updated tests for this change

Test details:

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code where necessary, particularly in hard-to-understand areas
• My changes generate no new warnings or errors
• I have updated the documentation accordingly (if applicable)

Additional Context

[dkosowski87]

Let's move the binarization_threshold to a constant, as other INFERENCE_MODELS_YOLO_ULTRALYTICS_DEFAULT_... ones

[PawelPeczek-Roboflow]

the other are configurable - does it make sense since the sigmoid is enabled?

Maybe it's actually valid to expose a knob to enable/disable smoothing and then this threshold being configurable makes more sense (on non smoothed distribution, cutoff point makes much more expression power)