Integrate Offline DAC Interface

In his thesis, Jannis wrote a logger for DE runs and parsing into D4RL-style datasets so we can train offline RL algorithms from unifloral. This would be a generally really nice feature to have, so we should integrate this functionality. Proposal:

- [ ] Add an abstract class based on the DE DAC Logger
- [ ] Use the env.reset() and env.step() functions in the Logger
- [ ] Append dataset parsing immediately after the Logger runs
- [ ] Add example using this for training with unifloral

Related: Jannis used detpy as DE implementation which might be nice to add as its own env since it has several dynamic baselines (and we already have the datasets). He mentioned it should be doable, but it not fully trivial for the baselines. Might need investigation.

Code: https://github.com/automl-private/Bsc-JannisKastner-DeDAC/tree/main

Start with the linked DeDAC implementation and trace how its logger uses env.reset() and env.step(). Compare that design with the proposed abstract logger, dataset parsing, and unifloral training example; done means all four proposal items are integrated, with detpy and its dynamic baselines investigated separately.

Written by the indexing model from the issue text.