[Breaking] Unify states modifications

[MartinuzziFrancesco]

At the moment we have nonlinear algorithms and padding/extension as separate entities although they both manipulate the states. Since not all manipulations neatly fit into these categories, we should unify the alterations to the states.

[MartinuzziFrancesco]

#238 is a first step in this direction.

Ideally we should create something like StatesChain that provides an unified and flexible interface to have any kind of modification possible:

states_mod = StatesChain(
    NLAT1(),
    PaddedStates(1.0),
    ExtendedStates(),
)

Maybe the naming of the states function will need some revisioning, but feels like the correct direction.

The only problem here is given by extending the states. This StatesChain should go after, for example, ESNCell and have as input only the output of the cell (or wrapper of it). If we want to have an unified flow of data through the high level ESN then accessing the input data for the extension is not straightforward.

[MartinuzziFrancesco]

No states chain, but building on #206 they would just be Lux layers that can be placed in any location of the Chain