Basic Vector Search Index

RBC NOMI stores documents as numeric embedding vectors. Given a collection of document embeddings and a query embedding, return the indices of the k documents with the highest cosine similarity to the query.

Formal Specification

Implement cosine_search(vectors, query, k), where:

1. vectors is a non-empty list of equal-length lists of real numbers.
2. query is a list of real numbers with the same dimension as every vector.
3. k is an integer between 1 and len(vectors).
4. The function returns a list of k integer indices, ordered from highest similarity to lowest.
5. If two vectors have equal similarity, return the smaller index first.
6. The cosine similarity is dot(a, b) / (||a|| * ||b||). If either vector has zero magnitude, define its similarity as 0.0.

The implementation should compute each similarity without modifying the input vectors. Focus on correctness and clear handling of zero vectors.