Introduce YOLOv8 models

[tfaehse]

• replace all YOLOv5 models
• always use inference size == training size
• show status in GUI (which stage the pipeline is currently in)
• rewrite blurring algorithm, optimizing memory and CPU footprint
• allow a few more output container types, support depends on system

[tfaehse]

Just a draft for now, a few things are missing:

• smaller models, the nano ones are still being trained
• bigger ones too? Medium gives almost no improvements for detection
• the blurring rewrite doesn't support the mask arguments yet
• requirements need a review