Proposal for likelihood specification and factory API

[tilmantroester]

This is a more-or-less working prototype of a yaml-based likelihood specification, following the discussion thread.

Some features I think are nice:

• Separation of model and data specification: specifying the model does not require knowledge about the data
• Flexible: by using jsonargparse for dependency injection, the model specification can use systematics directly from firecrown or elsewhere, without the factory having to know about these systematics.
• I tried to separate the yaml-parsing logic from the likelihood factory logic, so the yaml format is not a hard requirement

Some obvious things that are still missing

• Clusters and SN. This only supports 2pt at the moment. I'm not very familiar with clusters workflow and an earlier version I had has stopped working.
• Features such as scale and bin cuts, but that should be straightforward to implement.
• Robustness, testing, etc.

I hope this PR can function as a starting point of a discussion on where this specification might not be up to our tasks, potential implementation pitfalls, and other suggestions and comments.

[vitenti]

Hi @tilmantroester , we have been working on the redesign of two point in the last two meetings and we are using your work here as a source of ideas to the new design. Soon we will have a separate PR branch with our code to discuss.