main_task run validator does not look for missing turns

[isoboroff]

The run validator should look out for missing turns. Alternatively, the converter to trec_eval format should populate missing turns with a single bogus document.

By default, trec_eval averages over the intersection of topics (turns) in the run and the qrels. With the -c option, it averages over all topics in the qrels, but it still doesn't report scores for missing topics.