Preliminary work-around for solvation NaN atomic radii passing silently

This is a preliminary work-around for issue #53 (closed) (NaN atomic radii passing silently in solvation model). The radii are now checked explicitly to avoid erroneous calculations passing silently. A test has also been added. A proper fix for the root-cause (too generous masking of floating point errors) will follow.

I don't know yet, when I will have time to solve this properly, so I created a merge request for the "hot-fix". Issue #53 (closed) should stay open until this is fixed completely.