Refactor `ExpData`

[dweindl]

We might want to think about restructuring the current ExpData. In particular in the context of parameter estimation it would be great to split that into a parameter-independent and a parameter-dependent part.
The former, measurements / observations / dataset would be constant across the course of an optimization and only the latter, containing simulation parameters, potentially for multiple simulation periods, would have to be updated.
This way it would be easier to re-use and pass around the measurements in a memory-efficient way and only recreate/update the parameter-dependent part.

See also #931

For related bigger changes to the core simulation objects, see also #2845.