The unknown atom name error is not occuring when Rosetta is reading in your PDB file, it is occuring when other code in the program is trying to access the H atom on that residue. Proline doesn’t have a amide hydrogen, so trying to look it up is a pretty legitimate failure.
I think in both cases these failures are likely caused somewhere in the scorefunction. The standard scorefunctions make many assumptions about having normal protein; the scorefunctions used for noncanonicals (and Ds) do not. I’ve emailed the noncanonicals guy for advice on which scorefunction to use.