fix(gemini): handle thinking model response parts correctly

[mackenzieclark]

Summary

• Gemini thinking models (e.g. gemini-3-pro-preview) return response parts with thought: true (internal reasoning) and thoughtSignature (opaque signature blob). The current extraction logic takes the first part, which is the thinking part, returning reasoning text instead of the actual answer — or failing with "error decoding response body" when the response structure doesn't match expectations.
• Adds thought field to ResponsePart, implements effective_text() on CandidateContent to skip thinking/signature parts, updates extraction to use it.
• Makes Candidate.content optional to guard against candidates with no content (e.g. safety-filtered responses).
• 7 new inline tests covering thinking, non-thinking, fallback, empty, multi-part, signature-only, and internal API (cloudcode-pa) responses.
• Non-thinking models are completely unaffected.

Problem

When using gemini-3-pro-preview (or other thinking-enabled Gemini models), the API returns parts like:

{
  "candidates": [{
    "content": {
      "parts": [
        {"thought": true, "text": "Let me reason about this..."},
        {"text": "The actual answer."},
        {"thoughtSignature": "base64blob..."}
      ]
    }
  }]
}

The previous code blindly took .parts.into_iter().next(), which returned the thinking part instead of the answer. Parts with only thoughtSignature (no text) were noise that could cause empty responses.

Changes

src/providers/gemini.rs

• ResponsePart: adds thought: bool field (#[serde(default)]) to detect reasoning parts
• Candidate: makes content optional (Option<CandidateContent>) for robustness
• CandidateContent::effective_text(): new method following the effective_content() pattern from OpenAI provider — skips thought: true parts, skips signature-only parts, concatenates non-thinking text, falls back to thinking text if that's all that's available
• send_generate_content: uses effective_text() instead of .parts.into_iter().next().and_then(|p| p.text)

CHANGELOG.md

• Entry under [Unreleased] > Fixed

docs/providers-reference.md

• Note about thinking model support in Gemini Notes section

Validation Evidence

• cargo check --lib — compiles clean (no new warnings)
• cargo clippy — no issues in gemini.rs
• cargo fmt — no formatting issues in changed files
• Note: cargo test --lib is blocked by a pre-existing compilation error in src/agent/loop_.rs:3603 (missing 8th arg to build_system_prompt_with_mode) — unrelated to this PR

Security Impact

• Risk level: low
• Change type: additive response parsing only
• No new privileges, no policy changes, no secret handling changes
• Non-thinking models remain on the existing code path

Rollback Plan

• Revert this commit to restore previous extraction behavior
• No schema or config changes, no stateful data migration required

🤖 Generated with Claude Code

[github-actions]

PR intake checks found warnings (non-blocking)

Fast safe checks found advisory issues. CI lint/test/build gates still enforce merge quality.

• Missing required PR template sections: ## Validation Evidence (required), ## Security Impact (required), ## Privacy and Data Hygiene (required), ## Rollback Plan (required)
• Incomplete required PR template fields: summary problem, summary why it matters, summary what changed, validation commands, security risk/mitigation, privacy status, rollback plan

Action items:

1. Complete required PR template sections/fields.
2. Remove tabs, trailing whitespace, and merge conflict markers from added lines.
3. Re-run local checks before pushing:
   • ./scripts/ci/rust_quality_gate.sh
   • ./scripts/ci/rust_strict_delta_gate.sh
   • ./scripts/ci/docs_quality_gate.sh

Run logs: https://github.com/zeroclaw-labs/zeroclaw/actions/runs/22238834237

Detected blocking line issues (sample):

• none

Detected advisory line issues (sample):

• none

[github-actions]

Thanks for contributing to ZeroClaw.

For faster review, please ensure:

• PR template sections are fully completed
• cargo fmt --all -- --check, cargo clippy --all-targets -- -D warnings, and cargo test are included
• If automation/agents were used heavily, add brief workflow notes
• Scope is focused (prefer one concern per PR)

See CONTRIBUTING.md and docs/pr-workflow.md for full collaboration rules.

[chumyin]

Thanks for the contribution and collaboration.\n\nFinalization update:\n- Verified PR state, review status, and checks for merge eligibility.\n- Completed rebase merge.\n\nWe are currently conducting ZeroClaw automated testing. This is an automated comment from ZeroClaw. If you have any questions, please contact @chumyin.