Auto-identify bulk data as heavy or light chain

[willdumm]

Previously our only bulk data was heavy chain data, so we assumed that if the pcp file did not distinguish between heavy and light e.g. parent sequences, that we were talking about heavy chains. Now we have light chain data in the same format, and in theory we could have a pcp file with mixed heavy and light chain bulk data.

To handle this, we process pcp_dfs into a format that always contains heavy/light differentiated _h and _l columns, but we automatically infer the chain type for each pcp based on the v family name.
If the pcp file already has differentiated _h and _l columns, as with paired data, then we assume that no inference is necessary and only check for all necessary columns and make sure that all the heavy chain and light chain v families seem to be heavy or light, as claimed.

I also added a more informative error message for when masked parent-child nt pairs are identical, since I moved that filtering step to pre-processing in dnsm-experiments.