Describe the feature or idea you want to propose

all networks in aeon are parameterized, on number of layers and parameters per layers, default values as the ones in their published associated paper.
would be nice to have this option for mlp too, handling list inputs as for fcn and others

Describe your proposed solution

simply do for mlp network like for fcn network, and then add the arguments to the classifier and regresdor that uses the network

Describe alternatives you've considered, if relevant

No response

Additional context

No response