Add custom LM tasks to nextflow pipeline

Possibly also the data generation routines themselves. Why not?

Unfortunately the Books Corpus data is not publicly available, but we can at least provide this as an optional feature, given that the user provides their own input language modeling corpus.