Google Gemini adapter #34

Problem

AOBench cannot call Google Gemini. Its function-calling declarations differ structurally
from OpenAI's, so the existing openai adapter (which reaches any OpenAI-compatible
endpoint) does not cover it. That means no Gemini results on the leaderboard.

Desired result

aobench run task --task JOB_USR_001 --env env_01 --adapter gemini:<model>

Read this first

If the LiteLLM adapter (#33) lands, this may reduce to a documentation page plus a regression
test. Please check #33 before starting, and comment here so we do not do the work
twice.

Where to look

Implementation hints

• SDK: google-genai.
• The real work is the tool-schema translation. Gemini's function declarations are
  shaped differently from OpenAI's; the ToolRegistry schemas have to be converted, and
  the tool results have to be converted back into the Trace shape.
• Import the SDK inside the function; raise an actionable "install aobench[gemini]"
  error rather than letting an ImportError escape.
• Populate token counts — the efficiency scorer and CLEAR cost axis depend on them.

Acceptance criteria

• aobench run task --adapter gemini:<model> ... produces a scoreable Trace
• Tool calls and results recorded in the same shape as the openai adapter
• Token usage populated
• Missing dependency and missing credentials each produce their own clear message
• aobench list adapters includes it
• docs/guides/evaluating-your-own-agent.md updated

Tests

tests/unit/ with a mocked client, following the existing OpenAI adapter tests.
No network calls in tests.

Difficulty

Roughly a day.

Getting started

Comment before starting. Questions welcome.