Initialization of `SymbolicData` objects

[vincepandolfo]

Since #67 brought to light some issues with the initialization of SymbolicData objects (TimeData objects specifically) we need to find a coherent and uniform interface to initialize SymbolicData objects.

A possible solution for TimeData would be to have the data property reference _full_data instead of _data. In this way when self.initializer is called from initialize it will get as an argument the full data array instead of the data array with the hidden indexes.

This would also have the effect of removing the time padding from the user point of view.

Initial discussion with @mlange05 here: 09dc155#commitcomment-18517910

[mlange05]

I like the idea of having the data property referencing _full_data under the hood for time-buffered TimeData objects. I also note that this probably aligns with issue #69, where the obvious solution is to keep a private running timestep counter that we can modulo on to get the right time-frame in data.

Also, I believe the idea of adding the "initializer function" as a constructor argument was floated at some point, which I think sounds like a good idea? As does the idea of removing the pad_time argument? Just a few suggestions for further discussion...

[navjotk]

Agree on the initializer function and pad_time above.

[vincepandolfo]

Having data reference _full_data doesn't quite work since Operator.apply calls param.data, but it needs _data to work since when save=True the generated code tries to access the hidden indexes. A possible solution would be for Operator.apply to call param._data instead.