Add `HorizontalSMAModel`

- the model applies a simple moving average on horizontal time series
- refactor `db.Forecast.from_dataframe()` to correctly convert
  `float('NaN')` values into `None`; otherwise, SQLAlchemy complains