Lukas issue32 refac restructure plot routines in modules

See merge request toar/machinelearningtools!28

# Conflicts:
#	requirements.txt
#	src/run_modules/model_setup.py

added some docs, new plot routine plot_competitive_skill_score, skill score plots are now included in postprocessing, many changes and bugfixes in skill score calculation

bugfix: added start and end to the default kwargs keys, before the time range in all subsets (train, ...) have been equal!

added 4th subset "train_val" that consists on the union of train and val. Is needed for the external skill score calculation. Also added simple MSE function (no need to install sklearn just for mse)

first part of skill score calculation (CASE I and II implemented), added copy of skill_score_plot from felix repository (not checked yet)

minor change in data loading process, there can be now the option to re-download data during preprocessing

created detailed description of all data paths (which steps creates which data and where is it stored)

implemented test for history plot, monitoring plots adjust now if there is single or multiple output branches and if mse is available as metric

abstracted PlotModelHistory to be not only able to plot loss but also other metrics from the history callback object for a given plot metric

applied bug fix for #6, program can handle now networks with single or multiple output branches in the right manner

removed 1 failing test for postprocessing (module is not ready yet), another test sometimes fails because of some weird name changing (but this is only the name not the content)

# Conflicts:
#	requirements.txt
#	src/run_modules/experiment_setup.py
#	src/run_modules/model_setup.py
#	src/run_modules/training.py

post processing can load local model if no new training was performed and trained model was saved locally for the regarding experiment

post processing can plot station_map(), but countries are not filled with e.g. elevation data (just blank fill)

add plot and results path to experiment setup, finished raw code for prediction making in postprocessing