VertexAiSessionService creates a new vertexai.Client per call; SSL contexts accumulate on Agent Engine #7353

Describe the Bug:
VertexAiSessionService._get_api_client() creates a new vertexai.Client on every session call (create_session, get_session, append_event, ...). Each client creates two ssl.SSLContexts via google.auth.transport.mtls.get_default_ssl_context() (_ensure_httpx_ssl_ctx and _ensure_websocket_ssl_ctx), and they are not freed. On Vertex AI Agent Engine, where a client certificate is available, memory grows about 7 MB per request until the instance is killed.

Steps to Reproduce:

1. Deploy an ADK app to Agent Engine with get_fast_api_app(session_service_uri="agentengine://<engine id>").
2. Send a few hundred requests.
3. Watch process RSS and live ssl.SSLContext objects (gc.get_objects()).

Expected Behavior:
One API client is reused across session calls, and memory stays flat.

Observed Behavior:
About 9 live SSLContexts and about 7 MB of RSS are added per request, and none are released while idle. Grouped by creating stack, the largest groups (about 80% of the new contexts) all come from google/adk/sessions/vertex_ai_session_service.py:583 _get_api_client. Locally (no client certificate) it is smaller but still present: 12 session calls created 12 clients and left 3 extra live SSLContexts.

Environment Details:

• ADK Library Version: google-adk 2.10.0 (google-genai 2.25.0, google-cloud-aiplatform 1.165.1, google-auth 2.59.0)
• OS: Linux (Vertex AI Agent Engine)
• Python Version: 3.14

Model Information:

• Are you using LiteLLM: No
• Which model is being used: N/A (session service only)

Regression:
Possibly since google-genai 2.13.0 (googleapis/python-genai#2718), which made each client build an mTLS SSL context via get_default_ssl_context().

Minimal Reproduction Code:

import asyncio
import vertexai
from google.adk.sessions.vertex_ai_session_service import VertexAiSessionService

created = 0
_init = vertexai.Client.__init__

def counting_init(self, *args, **kwargs):
    global created
    created += 1
    _init(self, *args, **kwargs)

vertexai.Client.__init__ = counting_init

async def main() -> None:
    service = VertexAiSessionService(project="<project>", location="<location>", agent_engine_id="<engine id>")
    session = await service.create_session(app_name="<engine id>", user_id="u1")
    for _ in range(10):
        await service.get_session(app_name="<engine id>", user_id="u1", session_id=session.id)
    print("clients created:", created)  # 11: one per call

asyncio.run(main())