Currently values and variances of Variable and DataProxy returns a flat (1D) range of values.

• For better alignment with numpy and xarray, returning an object with correct dimensionality would be better. For this we currently have the numpy property.
• The distinction may have made sense historically in the design process, but definitely after the recent redesign of Dataset it does not seem to make much sense any more.

So, the suggestions/questions are:

• Remove the numpy property.
• Return N-D numpy-like-array from values and variances.
• Change the dims property to return only the labels (as xarray), get shape from values (or add a shape property to Variable and DataProxy).
• Should we also make this change on the C++ side?
• Any other thoughts/issues related to this?