[onert/train] Revisit nnfw experimental API

[YongseopKim]

Let's add to explain how to use nnfw experimental API and add a scenario.

Ref: https://github.com/Samsung/ONE/issues/12510#issuecomment-1905158219

IMO, we need to add comment on nnfw_train_prepare about this: when nnfw_input_size() and nnfw_output_size() is called after nnfw_train_prepare

Additionally, nnfw experimental API code could be changed for better usability.

/cc @jyoungyun @hseok-oh