Write DataFrameLike interface (either pandas or polars)

[t-ober]

We currently rely on the pd.DataFrame and pd.Series as the base data model. Rather than relying on these explicitly we could abstract that behind an interface e.g. DataFrameLike and SeriesLike which offer the same (or a subset of important) public methods.

This would allow us to easily switch out the actual data source for

1. a database interface (Implement SQL as possible data source #31)
2. a different data library (e.g. Polars Exchange pandas for polars for much better performance #28)

With respect to 2 we would need to consider if we can leverage the performance boosts of polars because some of them rely on a different of dealing with data. E.g. collecting tasks and the executing them in a combined fashion. Therefore the first task would be to find out if we should rather switch to polars first and do polars.DataFrameLike interface. Pandas would be much easier as we wouldn't have to change any of the implemented logic.