Feature Request: Add Milvus-backed Memory Service #69

Description

Add MilvusMemoryService, an implementation of BaseMemoryService that uses Milvus as the vector store for cross-session conversation memory.

Motivation

ADK's existing memory services (InMemoryMemoryService, VertexAiMemoryBankService, VertexAiRagMemoryService) either lack semantic search or require GCP infrastructure. Milvus provides a lightweight, open-source alternative that supports local (Milvus Lite), self-hosted, and fully managed (Zilliz Cloud) deployments.

This complements the MilvusToolset (PR google/adk-python#4417) which handles RAG knowledge base search, by adding a separate memory persistence layer for conversation history.

Proposed Solution

• MilvusMemoryService(BaseMemoryService): stores session events as vector-embedded text in Milvus, with app_name/user_id scoping and deduplication.
• Users provide their own embedding function (e.g., Google GenAI gemini-embedding-001).
• Works with PreloadMemoryTool and LoadMemoryTool out of the box.

Related PR

google/adk-python#4417