Implement a RAGApplication class that initializes a vector store, tokenizes text, and builds a vocabulary from a list of texts.

The RAGApplication class should maintain an in‑memory vector store and a simple vocabulary. In the constructor, initialize empty lists for documents and embeddings. The _tokenize method uses a regex to split text into lowercase tokens. The _build_vocab method iterates over the provided texts, tokenizes each, and updates a Counter to count token frequencies. After building the Counter, convert it to a list of unique tokens sorted by frequency to form the vocabulary. For embeddings, use a pre‑trained sentence transformer (e.g., sentence-transformers/all-MiniLM-L6-v2) to encode each document into a fixed‑size vector and store it alongside the document. Complexity: tokenization is O(n m) where n is the number of documents and m is average length; building the Counter is linear in total tokens; encoding each document is O(n d) where d is embedding dimension. The resulting class can then be used to retrieve the nearest neighbors by computing cosine similarity between a query embedding and stored embeddings, returning the top‑k relevant documents. This approach keeps the implementation lightweight while demonstrating core RAG concepts.