Improve index_names handling in ImpactDataInterpolator / rime-run-table

[perrette]

Currently rimeX.emulator.ImpactDataInterpolator.from_dataframe checks for a set of pre-defined fields: ['warming_level', 'year', 'region', 'model', 'variable', 'scenario'], and builds a n-D DataArray from that.

We should:

1. add --index and --meta option to rime-run-table to overwrite the default
2. in any case, build a 3-D (year not included) or 4-D (year included in the index) data-array, where the last dimension is flattened, instead of a n-D array. That would be less sparse, thus faster and likely yield better interpolation.

@mwerning @byersiiasa