Differentiate between observations and states

[thomaswolgast]

Currently, only the observation space gets defined. The same space is also used for sampling, which can result in side effects. For example, if the observation space is reduced (partial obs), the non-observed state channels might not sample correctly anymore.

Approach: Explicitly define state space for sampling and observation space for observations. Further things to to:

• Add two methods to return obs and state, respectively.
• Store state in info dict. (so that the algorithm can use the state as well)
• Open: How to deal with results? They require a costly power flow computation and are not strictly required for Markov. However, they are part of the power system's state.

Related to #10 and #8.

[thomaswolgast]

Do not store state in dict to save computation time. Maybe add this later, but then with popgym support (https://popgym.readthedocs.io/en/latest/autoapi/popgym/core/env/index.html).

Do not include results in state. They are not required to predict the next state, so they are not part of the Markov state.