Improve `ResultsHandler`

[mlincett]

Following up on #178 I would like to improve ResultsHandler.

As a first step, instead of using class ad-hoc named attributes to store sensitivity, discovery potentials and other outputs I would like to wrap all this in a single object.

Then I would like to add the calculation of the 3 sigma discovery potential (or at least make the significance threshold configurable).

Questions:

• should the result object have a flat structure (for example fields like sensitivity-value, sensitivity-error, discovery-potential-5sigma-value, etc.) or a nested structure (['sensitivity']['value'], etc.)? While the nested structure is more intuitive, it is also more complicate to implement and may not really make the object more usable;
• should we standardize the structure using native python dataclasses or be bold and open the door to the use of pydantic in flarestack (after @robertdstein suggestion)?

I will start by converting https://github.com/icecube/flarestack/blob/d93f5111baa4a3aa2eff0659d834a48cfa9cbac4/flarestack/core/results.py#L104-L114 to a dictionary but I would appreciate some feedback before moving onto more radical improvements.

[JannisNe]

I would really love to use a pydantic model. That way, it would also be really easy to export all the results as a json. I think it should not be that complicated to implement.

[JannisNe]

Re the nested-vs-flat question: I think a nested structure is not that much more useful or intuitive so I think a flat structure is better in terms of complication to usefulness.