milvus-io/milvus ? reverse-engineered prompt

Build me a high performance vector database that I can run locally or in a clustered setup, with a simple client flow for creating a collection, inserting embeddings, and searching for nearest matches. It should support text and image style vector data, let me filter by metadata, and return the top results fast even when there are lots of vectors.

I want a clean way to start it in standalone mode for development, but also make it ready for a cloud native distributed deployment with containers and Kubernetes when I need more scale. Please include sensible defaults, config files, and a straightforward Python client example so I can try it right away. If you need to look up current docs online, go ahead.

Also make sure the project feels production ready, with good docs, a working quickstart, and a clear path for adding different index types later like HNSW or DiskANN.