numpy-api: Enable support of Feols methods with numpy-api

[s3alfisc]

Context

Relates to #454. Currently, most post-estimation methods for inference require the input data frame to be stored within the model object.

Additionally, they might require context about the model formula, dropped columns etc.

When fitting a model via the OOP API, the Feols class will not have a _data, _fml, etc attribute. As a result, a range of methods need to be refactored.

To do

Refactor the following methods:

• vcov() needs to get a "cluster" argument, where users can provide a numpy array for the cluster variable. This array needs to be converted to ints internally.
• wildbootest() same as above, but no conversion to ints required
• ccv(): "param" and "cluster" should be changed to arrays
• [fixef()]() currently requires a formula on model inputs. need to convert fixef_array to one-hot encoded matrix without relying on fixed effects names.
• predict() the newdata likely needs two new arguments - new_covars and new_fixef.
• ritest(): resampvar, cluster should be allowed to be np.arrays

Additionally, the

• summary()
• etable()
• coefplot()
• iplot()

methods all require information on the model formula.

[s3alfisc]

Looks like rather high efforts ...