fix(google-genai): align Gemini 3 model ids with the ids Google serves

[erzaehlsalex]

Closes #209. Closes #210 (the per-model thinking levels landed in #215; this adds the remaining id and the per-id tests).

What

• Add gemini-3-flash-preview with its four thinking levels (minimal, low, medium, high). It previously fell back to the Gemini 2.5-style thinking-budget defaults.
• Remove dead ids from the capability table, tests and docs:
  • gemini-3-pro — the real id was gemini-3-pro-preview, shut down March 9, 2026.
  • gemini-3.1-pro — not an id on the Gemini API or Vertex AI; the model is gemini-3.1-pro-preview.
  • gemini-3.1-flash-lite-preview — shut down May 25, 2026; gemini-3.1-flash-lite stays.
• Fix examples in the docs, the root README and the package README to use ids that resolve (gemini-3.1-pro-preview, gemini-3-flash-preview).
• Tests now cover supportedEfforts for every Gemini 3 id in the table.

The Vertex question from #209

#209 asked whether the dateless ids are valid on Vertex AI. The Vertex model pages for Gemini 3.1 Pro and Gemini 3 Flash list gemini-3.1-pro-preview and gemini-3-flash-preview as the model ids, the same as the Gemini API. So the dateless gemini-3.1-pro entry is removed rather than kept, and no gemini-3-flash entry is added.

Behaviour

• gemini-3-flash-preview now sends thinkingLevel instead of thinkingBudget.
• Callers still passing a removed id get the default (thinking-budget) capabilities, as any unknown id does. Those ids no longer resolve at Google, so no working request changes.

Sources

• https://ai.google.dev/gemini-api/docs/deprecations
• https://ai.google.dev/gemini-api/docs/models
• Vertex model pages for Gemini 3.1 Pro and Gemini 3 Flash

Checks

• npm run release:check and npm run test pass.
• Live e2e tests were not run (no Google API key available locally).
• Changeset included for @core-ai/google-genai (patch).

🤖 Generated with Claude Code

Closes #209 and #210

[mtraute]

LGTM 👍