fix(csa-review): classify gemini quota-exhausted as rate-limit, not MCP (#871)

[RyderFreeman4Logos]

Summary

• detect_tool_diagnostic in review_cmd_output.rs previously matched only the literal "MCP issues detected" / "Run /mcp list" substrings. Gemini-cli error envelopes that contain BOTH the MCP hint AND the real QUOTA_EXHAUSTED cause were misclassified — users saw "Run /mcp list" guidance and an incorrect fallback suggestion (--tool claude-code) when the actual remedy is to configure GEMINI_API_KEY or wait for quota reset.
• Add a quota-first branch that reuses csa-core::gemini::RATE_LIMIT_PATTERNS (plus \"quota will reset\") and emits GEMINI_API_KEY-fallback guidance. The MCP branch still fires for genuine MCP-server connectivity issues.

Closes #871.

Test plan

• cargo test -p cli-sub-agent detect_tool_diagnostic
• cargo test -p cli-sub-agent review_cmd
• cargo test --workspace
• just pre-commit

🤖 Generated with Claude Code

[RyderFreeman4Logos]

Local pre-PR cumulative review (csa review --range main...HEAD, tier-4-critical, session 01KPNR0Z2RQ0JJ0YWKAVYHV1GC):

• reviewer 1 (gemini-cli): UNCERTAIN (no specific finding)
• reviewer 2 (codex): "No P0/P1 findings" + "Ship as-is". findings.toml: findings = [].
• Consensus token: HAS_ISSUES (parser FP — aggregates reviewer 2's prose-only advisory into a verdict despite empty findings list). Same review-verdict parser: decision=fail when reviewer summary explicitly says CLEAN but full.md lacks CSA structured findings section (post-#909) #946-class parser FP that's been documented in prior merge audit trails.

Only advisory concern raised: the reused RATE_LIMIT_PATTERNS set is shared with the scheduler's rate-limit detector, and a bare "429" substring in gemini review output would emit the quota diagnostic even if not truly a quota error. Triggering conditions are narrow (tool must be gemini-cli AND output contains literal 429), and the worst-case message is "configure GEMINI_API_KEY or wait for reset" — helpful-or-neutral, not incorrect guidance. Optional narrower matcher is a valid followup; tracked for future hardening.

Merging per severity-tiered protocol: findings=[] + 0 P0/P1 + both reviewers' prose directs "ship". No production regressions introduced — only a new diagnostic branch.

[gemini-code-assist]

Code Review

This pull request bumps the project version to 0.1.445 and introduces logic to detect Gemini API quota exhaustion within tool output. The detect_tool_diagnostic function now identifies rate-limit patterns and provides a diagnostic message suggesting the configuration of an API key. New unit tests verify this detection and ensure quota issues are prioritized over MCP-related errors. Feedback was provided to improve the diagnostic message by removing a hardcoded Unix-style file path and clarifying the distinction between environment variables and configuration fields.

[gemini-code-assist]

The diagnostic message hardcodes a Unix-style path (~/.config/cli-sub-agent/config.toml) which may be incorrect for users on other platforms (e.g., Windows). Additionally, the wording "configure GEMINI_API_KEY ... under ... api_key" is slightly confusing as it mixes the environment variable name with the configuration field name. Consider using a more generic reference to the configuration file and clarifying the field name.

Suggested change

	return Some(
	"gemini-cli OAuth quota exhausted. Either (a) configure GEMINI_API_KEY in ~/.config/cli-sub-agent/config.toml under [tools.gemini-cli] api_key for automatic retry, or (b) wait for quota reset."
	.to_string(),
	);
	return Some(
	"gemini-cli OAuth quota exhausted. Either (a) configure your Gemini API key in the CSA config file under [tools.gemini-cli] api_key for automatic retry, or (b) wait for quota reset."
	.to_string(),
	);