Integrate not-stolen image generation for target matching

[atar13]

🚀 Feature

Motivation

The way our target matching model works is that it needs reference images of all the competition objective targets to compare against. For example, if the competition tells us to look for a blue triangle with a white "A" then we should generate a few variants of targets under that description for our matching algorithm to compare against. We want to generate multiple targets to have some variability (different hues,saturation, etc.)

Pitch

Port over functionality from not-stolen to the obcpp to allow use to generate cropped targets based on a given shape type, shape color, character type, and character color. This is similar to how you can specify a subset of targets to generate in not-stolen https://github.com/tritonuas/not-stolen-israeli-code/blob/master/exampleSubset.yaml.

Alternatives

Instead of porting over the code to C++, we could call our existing Python implementation via the shell and just read the target images that it outputs.