Cleanup typing of internal `X, y` data passed around

[eddiebergman]

Currently we accept the following kinds of data and use the type internally throughout or selectively just use one of them when more than one is appropriate. Generally, there is also quite a few mypy errors for not covering all cases of type signatures which are too general.

SUPPORTED_FEAT_TYPES = Union[List, DataFrame, ndarray, spmatrix]
SUPPORTED_TARGET_TYPES = Union[List, Series, DataFrame, ndarray, spmatrix]

Firstly, due to the general lack of support in sklearn for y's being spmatrix, we should probably remove that, keeping the check and issuing the warning about densification if a sparse y is passed in.

Second, a proposed change to have the following types:

X_INPUT_T = Union[List, DataFrame, ndarray, spmatrix]
Y_INPUT_T = Union[List, DataFrame, ndarray, Series]
 
XT = Union[ndarray, spmatrix]
YT = ndarray

The X_INPUT_T and Y_INPUT_T are types exposed to the user interface but the input validator converts all data:

class AutoML:
    
    def fit(X: X_INPUT_T, y: Y_INPUT_T, ...):
        ...
        X_transformed : XT
        y_transformed : YT
        X_transformed, y_transformed = input_validator.transform(X, y)
        ...
        # Use X_transformed and y_transformed with type signatures of XT and YT from here        

This would mean updating the estimators interfaces as well

Bonus points for a function that specifies that it returns the same type it gets in:

T = TypeVar("T", ndarray, spmatrix)
def process(X: T) -> T: ...

[eddiebergman]

Further info suggets that sklearn disregards any information from pandas dataframes, highlighting the fact we should convert dataframes to ndarrays internally, making our types to only be spmatrix and ndarray
https://scikit-learn.org/stable/faq.html#why-does-scikit-learn-not-directly-work-with-for-example-pandas-dataframe

Whoops, this is done as soon as the data goes through the InputValidator. However, we should have types to indicate transformed version of the data

[eddiebergman]

Following PEP8 standards of short names, it's probably just best to use XT and YT
https://www.python.org/dev/peps/pep-0008/#type-variable-names