[ENH] clean-up refactor of TimeSeriesDataSet

[fkiraly]

This PR carries out a clean-up refactor of TimeSeriesDataSet. No changes are made to the logic.

This is in preparation for major work items impacting the logic, e.g., removal of the pandas coupling (see #1685), or a 2.0 rework (see #1736).
In general, a clean state would make these easier.

Work carried out:

• clear, and complete docstrings, in numpydoc format
• separating logic, e.g., for parameter checking, data formatting, default handling
• reducing cognitive complexity and max indentations, addressing "code smells"
• linting