test(agent-loops): cover claude/gemini provider diagnostics in review scripts

[qiaobochi040726-source]

Fixes #29

Summary

Adds explicit Claude and Gemini diagnostic-message test coverage to both agent-loops review entrypoints:

• specialist-review.sh — 2 new tests: explicit Claude run verifies Claude budget + Claude model override diagnostics; explicit Gemini run verifies Gemini model override; each case asserts diagnostics from inactive providers are absent.
• diff-test-audit.sh — same 2 new tests. This script was missing the Claude model override stderr diagnostic entirely (it only printed Claude budget), so the fix adds the same 3-line block specialist-review.sh already has, enabling the Claude test to assert it.

Changes

• skills/agent-loops/scripts/diff-test-audit.sh: emit Claude model override: <model> when CLAUDE_MODEL is set (mirrors specialist-review.sh lines 555-559).
• tests/unit/test_agent_loops_review_scripts.py: +330 lines, 4 new tests (test_specialist_review_reports_claude_provider_diagnostics, test_specialist_review_reports_gemini_provider_diagnostics, test_test_review_reports_claude_provider_diagnostics, test_test_review_reports_gemini_provider_diagnostics).

Acceptance criteria

• Explicit Claude runs verify the configured model/effort diagnostic in both entrypoints
• Explicit Gemini runs verify the configured model diagnostic in both entrypoints
• Each case verifies diagnostics from inactive providers are absent
• Focused provider-script test suite passes (CI)

Note: execution behavior was already covered and is untouched; this change is limited to operator-facing stderr diagnostics and their tests.