Sub tick predictions for COB on sparse data

[Kalhama]

We do linear interpolation on known COB values, which might not represent the reality. If tick is too long (data is sparse) we should add sub tick predictions in the following matter.

Essentially we need to run the existing prediction function but for active meals at tick_start until tick_end at the rate attributed_carbs implies.

[Kalhama]

This issue is somewhat blocked by: #129

Before that is fixed the subticks won't yield relevant results.

[Kalhama]

Should we look back or forward with the rates?