[TEP002] Final changes to the various to_hdf methods

[ftsamis]

Note: When this is merged it will mean that the upstream to_hdf branch is ready to be merged with master.

This basically introduces a io.util.to_hdf method, as a uniform way to store various properties to HDF.

Currently the rules by which properties are stored are the following:

1. Scalars for each module (plasma,model,etc) are stored in a Series under <path>/scalars
2. 1D arrays are stored under <path>/property_name as distinct Series
3. 2D arrays are stored under <path>/property_name as distinct DataFrames
4. astropy Units are stored as their SI value
5. As an exception to 2 and 3, the runner, when stored in an HDF file, will have an arrays value which will contain the 1D arrays, instead of having a separate Series for each one of them. Don't ask me why, I don't know.
6. There is a special case for model.plasma.levels which is of type pandas.MultiIndex which in turn if passed to a pandas.Series for storage throws a NotImplementedError. Therefore I store it as a pandas.DataFrame which in turn produces a PerformanceWarning when calling to_hdf on it.