Refactor/remove protocol classes

[selmanozleyen]

Solves #765

There are some cons atm:

• I used Any annotations in the abstract classes which is less specific than most of the protocols. See for example:

class AbstractSolutionsProblems(ABC, metaclass=CombinedMeta):

    @property
    @abstractmethod
    def solutions(self) -> Any:
        """Solutions."""
        pass

    @property
    @abstractmethod
    def problems(self) -> Any:
        """Problems."""
        pass

    @property
    @abstractmethod
    def _policy(self) -> Any:
        """Subset policy."""
        pass

• For example one class uses src vs tgt terminology ( and other uses sp vs sc. They are essentially the same but I just rewrote them as is as two abstract classes. Otherwise it might also require more time to unify them.
  These are the two classes I am talking about:

class AbstractSrcTgt(ABC, metaclass=CombinedMeta):

    @property
    @abstractmethod
    def adata_src(self) -> AnnData:
        """Annotated data object."""
        pass

    @property
    @abstractmethod
    def adata_tgt(self) -> AnnData:
        """Annotated data object."""
        pass


class AbstractSpSc(ABC, metaclass=CombinedMeta):

    @property
    @abstractmethod
    def adata_sp(self) -> AnnData:
        """Annotated data object."""
        pass

    @property
    @abstractmethod
    def adata_sc(self) -> AnnData:
        """Annotated data object."""
        pass

• With the GENOT implementation GENOTLinProblem tries to inherit a class that requires push/pull
  

  moscot/src/moscot/problems/generic/_generic.py

  Line 779 in f84f8be

  class GENOTLinProblem(CondOTProblem, GenericAnalysisMixin[K, B]):

  
  pros:

• Less ignore comments for mypy

• Less code

• Better understanding of the expected attributes and methods

Things to do:

• Decide which files these new abstract classes should be in final version.
• Maybe fix the cons if important

[MUCDK]

This is amazing, thanks @selmanozleyen