split config file for creating hpf.csv from the rest of the pipeline

[mmaiers-nmdp]

Deconstructing py-graph-imputation in to three stages is easier if the config file is split into two
generate-hpf.conf.json - needs to know where frequencies are and where to put the hpf.csv
impute-graph.conf.json - needs to know where the hpf.csv is and the subjects to impute. Don't split graph creation from imputation conf files: they need to use the same PlanB_Matrix and locus arragement.