Build examples to enable people to train potentials sans n2p2 and use all the "rascal goodies"

[rosecers]

[rosecers]

Could include recreating previously trained potentials with rascal

[max-veit]

Right now we're thinking of three notebooks:

• [ ] Update and document MLIP_example.ipynb (see #314 )
• [ ] Running with i-PI (and eventually LAMMPS)
• [ ] Optimizing a potential for speed (e.g. feature sparsification, radial splining and contraction)