Full Observation -> Detection Pipeline

To speed up the processing the detection is happening outside of the game engine on a pre-parsed dataset.

This is fine in most cases!

But to have total freedom over the data. The pipeline should be as follows:

1. Run a full observation of all of the gameloops where players made an action.
2. Run the detection and prepare Intervals.
3. To do anything with the data query the specific detector, get its intervals, and return observations that are within the `(start_interval, end_interval)`

This will allow for comprehensive segmentation of the in-game data by various methods and loading different parts of the game for separate analyses.