Testcount forecasting

[lhelleckes]

This PR adds a module data_preprocessing.py which allows to forecast testcounts using fbprophet.

• retrieve country- and region-specific holidays to include in forecasts
• predict testcounts based on historic data with either a linear or a logistic growth model
• predict historic outliers by replacing them with numpy.nan
• example notebook to demonstrate use-case for US data (https://github.com/lhelleckes/covid-model/blob/9d1fe5ded234336a140c26f795787f4cabcb1778/notebooks/Example_Testcount_preprocessing.ipynb)

Since the forecast can also be applied for historic data, it serves as a smoothing for anomalies (#8). For regions with delayed data publishing, it can be used to forecast the missing data.

[michaelosthege]

With the latest changes, the get_holidays function became more flexible.
Most importantly with predict_testcounts(region="all", ...), testcounts are forecasted while taking into account all national and all regional holidays.

Also, a bug was fixed such that now the function can actually predict holiday-induced drops in testcounts.

[canyon289]

Nit small typo