[Bug]: /api/v2/health answers healthy while the vector store is unreachable #1777

Describe the bug

GET /api/v2/health returns 200 {"status": "healthy", ...} when the vector store cannot be reached, so a server that cannot store or search episodic memory is reported as healthy. Seen on main at ad8ff24.

The handler returns a constant and checks no dependency (packages/server/src/memmachine_server/server/api_v2/router.py:1097):

@router.get("/health", description=RouterDoc.HEALTH_CHECK, tags=["System"])
async def health_check() -> dict[str, str]:
    """Health check endpoint for container orchestration."""
    return {
        "status": "healthy",
        "service": "memmachine",
        "version": get_version().server_version,
    }

Its docstring says it is for container orchestration, and the repo's own docker-compose.yml uses it as the MemMachine container's healthcheck:

healthcheck:
  test: ["CMD", "curl", "--fail", "--silent", "http://localhost:8080/api/v2/health"]

So Docker, a load balancer, or a Kubernetes probe pointed at this endpoint keeps the instance in service while its episodic adds and searches fail.

Steps to reproduce

1. Start MemMachine with the event-memory backend on PostgreSQL and Qdrant (docker-compose.yml, Qdrant profile).
2. Confirm it is healthy and works: GET /api/v2/health returns 200, and an add and a search succeed.
3. Make Qdrant unreachable without closing its port: docker pause <qdrant container>.
4. Call GET /api/v2/health. It answers 200 {"status": "healthy", ...}.
5. Call POST /api/v2/memories/search on an existing project. It fails.