Add gemini-3.5-flash on google backend

[lchoquel]

Summary

• Adds gemini-3.5-flash to the google backend (adaptive thinking, text/images/pdf in, structured outputs, costs { input = 1.5, output = 9.0 } per million tokens).
• Registers it in the google LLM test collection.
• CHANGELOG updated under [Unreleased] > Added.

Test plan

• make tb — boot test passes
• make test-inference-fast PROF=test_gemini35flash_google TEST=TestLLMInference — 5/5 pass on real Google API
• make agent-check — lint + pyright + mypy clean
• Gateway-side validation pending pipelex-back-office deploy

🤖 Generated with Claude Code

---

Summary by cubic

Adds gemini-3.5-flash to the google backend for adaptive multimodal inference and structured outputs. Exposes the model in configs, tests, and Pipelex Gateway docs.

• New Features
  • Inputs: text, images, PDF; outputs: text, structured; max_prompt_images = 3000.
  • Costs { input = 1.5, output = 9.0 } per million tokens; thinking mode adaptive.
  • Registered in the google LLM test profile; listed in Pipelex Gateway model tables and plain list; noted in CHANGELOG.md.

^{Written for commit 08ff3dd. Summary will update on new commits. Review in cubic}

[greptile-apps]

Greptile Summary

This PR adds Gemini 3.5 Flash support for the direct Google backend. The main changes are:

• Adds gemini-3.5-flash to both Google backend config copies.
• Enables text, image, PDF, and structured-output capabilities for the new model handle.
• Registers the model in the Google LLM test collection.
• Documents the new model in the changelog.

Confidence Score: 5/5

This looks safe to merge.

• No blocking issues found in the changed code.

_{Reviews (1): Last reviewed commit: "feat: add gemini-3.5-flash model to Pipe..." | Re-trigger Greptile}

[cubic-dev-ai]

No issues found across 4 files

_{Re-trigger cubic}