Build Top-K Semantic Search

Implement top_k_search(query, documents, k): a list of (index, score) pairs for the k most similar documents, best first. cosine_similarity from the previous problem is already defined for you.

1. Sort by score descending, ties broken by the lower index.
2. If k exceeds the number of documents, return all of them. Don't pad, don't raise.
3. k <= 0 returns an empty list.
4. Scores are plain floats.