Previously forecast offset `df_scenario` was computed only when number of scenarios was
greater than one. Long-term scenario data should use a special label in the `f` dimension
in all cases.

This reverts commit b3a87bcf

Probabilities of single scenarios can be slightly above one as returned
by SCENRED2. Limit the probabilities to one.

Added new parameter `ts_energy_(s)` to which total energy is
calculated. This parameter has the sample dimension last, as required
by the SCENRED2 tool. This allowed changing all other time series
to have indices `s, f, t` instead of previous `f, t, s` and dropping
the set `fts` completely.

Fixes #105

Removed now obsolete scalar `errorcount`.

Solves #102

Fixes #100

Set `gn_forecasts(*, node, timeseries)` tells for which grid/flow/restype, node and timeseries combinations use forecasts and by default includes everything.

See: #99

#82: All hard-coded *elec* grids have been removed and replaced by looping/summing over *gn* instead. This fix should work as long as no *node* belongs to multiple *grids*, which I think is already ill-advised (or even forbidden?).

Also a minor speedup for setting model dimension (set `msft` etc.)

Changed behaviour of `mSetting` 't_perfectForesight' so that it’s defined from zero (not after 't_jump')

The setting does not have any effect if 't_perfectForesight' is less than 't_jump'.

Parameter `p_msft_probability` was not updated after scenario reduction.

Default is zero, can be changed via command line parameter --debug=[0|1|2].

Use --input_dir=<path> and --output_dir=<path>. Defaults are ./input and ./output, respectively.