[ENH] speculative design for raw data container DataSet v2

[fkiraly]

This PR contains a speculative design for a data container TimeSeries.

This follows the idea to split the current TimeSeriesDataSet in two parts:

• a "raw data set" object, representing a collection of time series or a single time series - contained in this PR as docstring
• a "resampled data set" which is based on a raw data set and additional resampling instructions - not contained in this PR

The idea of a split was discussed in #1736, but also already occurs in the comments of @jdb78 in the current TimeSeriesDataSet.

Iteration on #1755 upon feedback of @thawn that the output should be simpler, as few tensors as possible.

This design returns a near-minimal set, making the choice to:

• separate endogenous from exogenous
• separate future from past data
• otherwise store metadata as list of str or int

This is near-minimal, as at least two tensors are needed, and this results in three. Two tensors are needed since the data is the union of two rectangular array, but not itself a rectangular array (because some features are not seen in the future). The "future features" are optimal though.