Bugfix for #33

This is the bugfix of #33 (closed). The reason for the original bug was a typo, where deepcopy of the polynomial worked properly, but the wrong version of the polynomials was returned.

However, fixing that resulted in a follow-up bug in Lagrangian add. The origin of this follow-up bug was located in the expand_dim member function on MultivariatePolynomialABC, where the Grid instance associated with the respective polynomial was not expanded, if the respective multi_index was. To fix that, the _expand_dim utility function has been updated, and the grid was expanded to higher dimensions as well by filling the additional axes with zeros. This mimics the fact, that the (lower dimensional) grid is embedded in a higher dimensional grid-space, where the original grid is pinned to the origin of the higher-dimensional space.