chore(deps): update dependency google-genai to v2.23.0

[knostic-renovate]

This PR contains the following updates:

Package	Update	Change
google-genai	minor	==2.22.0 -> ==2.23.0

---

Release Notes

googleapis/python-genai (google-genai)

v2.23.0

Compare Source

Features

• Add from_environment support for environment copying (e6d3168)
• Support dynamic turn completion via interaction_status in live.py receive() (455df80)

Bug Fixes

• Remove unused field from FunctionResultDelta (2843079)

---

Configuration

📅 Schedule: Branch creation - At any time (no schedule defined), Automerge - At any time (no schedule defined).

🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.

♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.

---

• If you want to rebase/retry this PR, check this box

---

This PR has been generated by Renovate Bot.

[gadievron]

Verification review — recommend-merge. (head cfd43f8; evidence pinned to it)

Scope: single-line bump — google-genai==2.22.0 → ==2.23.0.

Upstream facts (re-derived from PyPI + the upstream changelog): requires_dist byte-identical 2.22.0↔2.23.0 (18 deps, including httpx>=0.28.1,<1.0.0 unchanged — the adapter's legacy-httpx catch net stays valid); requires_python >=3.10; not yanked; nothing beyond 2.23.0. Changelog: additive (from_environment for env copying; dynamic turn completion via interaction_status in live.py receive()) — the adapter only calls models.generate_content; the additions are unreachable.

Probes (fresh venv at the head, 6/6): str(FinishReason.STOP/MAX_TOKENS) forms UNCHANGED across both versions (re-derived in two environments: 'FinishReason.STOP' / 'FinishReason.MAX_TOKENS', matching the adapter's _GEMINI_FINISH_REASONS keys exactly — the enum-__str__ drift hazard closed); ClientError positional construction carries .code; HttpRetryOptions(attempts=...) declared; hermetic genai.Client construction; the GoogleAdapter constructed with a real client and a conn-refused completion mapped to LLMConnectionError — over the SDK's DEFAULT transport (no injected client; base_url via HttpOptions to an unroutable port), so the mapping holds on 2.23.0's own transport stack. Tests: the SDK contract floor + the google adapter suite — 37 passed. CI at the head: tests + gitleaks success.

Follow-up (pre-existing, named): the in-repo test factory raises its own transport errors, so the suite alone cannot detect a future default-transport flip — the real-transport conn-refused probe above covers this bump; a permanent non-stub guard would live in the factory.

Approve + merge are code-owner actions; never on red CI.