Consistency in coefficients for Stevens expansion

[kbarros]

The print_site function accepts an optional parameter R that sets a reference frame for the Stevens expansion. Currently, the coefficients ($c_1$, $c_2$, ...) are calculated from scratch for each new reference frame. Instead, we could give the coefficients a consistent meaning by first calculating the Stevens expansion for R=I and then rotating the expansion symbolically.

We could also maintain consistency between two sites by allowing to select a reference atom i_ref, which would be analogous to the b_ref parameter in print_bond.