[API] Centralize common operations that are used to make Context objects initialize state for learning algos

[adam2392]

Currently, for algo.fit(data, context), the data is a pandas dataframe that encodes the dataset for a single domain (or a list of data frames for a multi-distribution setting) and context is a Context object that contains a WIP API to enable causal-learning.

The issue is that it is not exactly clear how we can implement the Context API such that it is modularizable across different learning algorithms. For example, this is already coming up in #129, where some of the node names were being dropped because there is no common way for Context to be used.

We should centralize an abstract BaseLearner for all classes, which maybe initializes state in some way and has a common API that users and developers can rely on. For example:

class BaseLearner:
     # main APIs
     context: Context
     init_graph: Graph
     causal_variables : Set[Node]
    
    def _initialize_learner(self, data, context):
          # which should be called for every learner inside `def fit()`.
          # maybe also handle error-checking

This is a WIP issue to make sure #129 and future PRs can proceed robustly.