Feature Request: Add Milvus-backed Memory Service

### Description

Add `MilvusMemoryService`, an implementation of `BaseMemoryService` that uses [Milvus](https://milvus.io/) 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

Start by reading BaseMemoryService and the existing InMemoryMemoryService, VertexAiMemoryBankService, and VertexAiRagMemoryService implementations. Use the proposed MilvusMemoryService behavior as the scope: vector-backed session events with app_name/user_id scoping and deduplication, compatible with PreloadMemoryTool and LoadMemoryTool. The issue does not name specific files or tests, so the project’s existing memory-service conventions need to be located first.

Written by the indexing model from the issue text.