feat(memory): Add QdrantMemoryService backed by Qdrant vector search #193

Summary

Add a QdrantMemoryService to google.adk_community.memory: a BaseMemoryService backed by Qdrant vector search, scoped per app_name and user_id.

Motivation

Core ADK ships InMemoryMemoryService (keyword match, not persistent) and the Vertex AI services (GCP only). Self-hosted and non-GCP users need a persistent, semantic memory backend. Qdrant is a widely used open-source vector DB that runs as Docker, Qdrant Cloud, or in-process (AsyncQdrantClient(":memory:")), so one service covers local dev, CI and production. There is no Qdrant integration in adk-python or in this repo today.

Proposed API

from qdrant_client import AsyncQdrantClient
from google.adk_community.memory import QdrantMemoryService, QdrantMemoryServiceConfig

memory_service = QdrantMemoryService(
    client=AsyncQdrantClient(url="http://localhost:6333"),
    embedding_function=my_embed,  # async (list[str]) -> list[list[float]]
    config=QdrantMemoryServiceConfig(collection_name="adk_memory"),
)

• The caller owns the client and the embedding function, so the package doesn't depend on any embedding provider (same signature as the Valkey PR feat: add ValkeyMemoryService with vector similarity search (Valkey Search module) #156).
• Implements add_session_to_memory, add_events_to_memory, add_memory, search_memory and delete_memory.
• Point ids are deterministic (app/user/event id), so re-adding a session upserts instead of duplicating.
• Every search and delete is filtered by app_name and user_id, so one user can't read or delete another user's memories.
• The collection and payload indexes are created on first write (opt-out).
• New optional extra qdrant = ["qdrant-client>=1.12.0"], imported lazily, so the base install doesn't change.

Testing plan

• Unit tests with a mocked AsyncQdrantClient (collection creation, payloads, scoping, ids, error paths).
• One test against qdrant-client's in-process mode with a fake embedding. It needs no server, so it runs in CI.
• Manual E2E with a Runner and a Qdrant Docker container; logs will be in the PR.

I have the implementation and tests ready and will open a PR linked to this issue, followed by a docs page in google/adk-docs.

I will work on this.