Improve implementation of convert_dtypes.

[ioannis-vm]

Currently the line .fillna(value=np.nan) does not work well for dataframes that contain columns with mixed types, and pd.to_numeric(x, errors='ignore') is subject to deprecation.
We need to update our implementation to something more hands-on, converting the types and filling nan values column-by-column.

pelicun/pelicun/base.py

Line 883 in 89a13ae

def convert_dtypes(dataframe):