Support hyper-parameter search for next-item recommenders (#643)

[nsiddharth]

Description

GridSearch and RandomSearch (cornac.hyperopt) could not be used to tune NextItemRecommender models evaluated with NextItemEvaluation. Two issues blocked it:

1. The search wrapper was rejected by the evaluator. NextItemEvaluation.evaluate() requires its model to be a NextItemRecommender, but a GridSearch/RandomSearch object is a Recommender wrapper, not a NextItemRecommender. It raised:

   ValueError: model must be a NextItemRecommender but '<class 'cornac.hyperopt.RandomSearch'>' is provided
   

2. The search loop scored next-item models with the wrong evaluator. Inside BaseSearch.fit, ranking metrics fell through to the standard ranking_eval, whose rank()/score() path is incompatible with the session-based signature score(user_idx, history_items, ...):

   TypeError: SPop.score() missing 1 required positional argument: 'history_items'
   

Fix:

• NextItemEvaluation.evaluate() now also accepts a search wrapper whose .model is a NextItemRecommender.
• BaseSearch.fit() routes NextItemRecommender candidates through next_item_evaluation.ranking_eval, passing the eval method's exclude_unknowns and mode.
• BaseSearch delegates transform/score/rank to the selected best_model, so the fitted wrapper evaluates transparently on the test set.

This is additive — the existing rating/standard-ranking search paths are unchanged (verified by the pre-existing test_grid_search/test_random_search).

Related Issues

Fixes #643.

Checklist:

• I have added tests.
• I have updated the documentation accordingly.
• I have updated README.md (if you are adding a new model).
• I have updated examples/README.md (if you are adding a new example).
• I have updated datasets/README.md (if you are adding a new dataset).