Calculate recall (eval_metric) for xgboost

It would be nice to have `recall@k` as an evaluation metric. Currently, we can only use `ndcg`, `map` and `auc` for evaluating learning2rank problems.

Start by examining how the existing ndcg, map, and auc evaluation metrics support learning-to-rank problems. Determine the expected recall@k behavior and integration points; the work is done when recall@k can be selected as an evaluation metric and is covered by the relevant evaluation checks.

Written by the indexing model from the issue text.