Add api_version field to Gemini to support stable Vertex AI v1 API - #6711

ADK currently hard-codes v1beta1 as the Vertex AI API version for all Gemini
requests, including both generate_content and live-connect paths. This means
production workloads running on GA models (e.g. Gemini 2.5 Pro) are still
hitting a beta endpoint, which is outside the Vertex AI Gemini SLA.

This PR adds an api_version field to the Gemini model class. When set it
overrides both the version inferred from a base_url suffix and the
per-backend defaults (v1beta1 for Vertex AI, v1alpha for Gemini API). The
field flows automatically into api_client, _live_api_client, and every
generate_content / connect call via the existing _base_url_and_api_version
/ _live_api_version machinery.

Users can opt into the stable endpoint with:

from google.adk.models import Gemini
agent = Agent(model=Gemini(model="gemini-2.5-pro", api_version="v1"))

Three regression tests are added to test_google_llm.py covering the override
at the live-version, base-url-conflict, and api-client levels.

Fixes #3246