Carb predictions

[Kalhama]

Rewrite carb predictions

• for sparse (observed carbs) data, use static predictions
• for dense data use dynamic predicitons

static prediction

• use user reported rate and decay for prediction

dynamic predictions

• for first 15-30 mins of meal, use static prediction
• after that use recently observed rate
• gap the results to if they get too far from reported values

---

Simplification: This is a simplification to what we used before. I think it does not make sense to write complex carb attribution logic for simultaneous meals, because it will anyway. With simple we allow users to explicitly adjust meal parameters if the predictions go wrong. If we use complex logic users don't understand how it works and don't know how to do that.

Bug fixes: Current release of carb prediction does not work properply. Decay is too fast and carb plotting is broken

Maintainability and refactoring: simpler design allows to simplify the code