feat(cli): support compound tier-tool selectors (#1441)

[RyderFreeman4Logos]

Summary

• Parse compound --tier tier-4-critical-codex as tier=tier-4-critical + tool=codex, iterating hyphen splits from rightmost to handle multi-segment tool names
• Extract tier helper methods (try_parse_compound_tier_tool, suggest_tier, format_tier_aliases) into config_tier_helpers.rs to keep config.rs under 800 lines
• Extract test barrel from review_cmd.rs into review_cmd_tests_barrel.rs to stay under monolith gate
• Wire compound parsing into run_cmd with proper precedence: compound tier-tool overrides auto-select but explicit --tool still wins
• Add comprehensive tests for compound parsing, suggestion, and alias formatting

Test plan

• cargo test -p csa-config — tier selector tests pass
• cargo test -p cli-sub-agent — review_cmd tests pass via barrel
• just pre-commit — all gates pass
• CSA codex review verdict: PASS

Closes #1441

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request implements support for compound tier selectors in the format --tier <tier>-<tool>, enabling users to specify a tier and a tool override simultaneously. The implementation includes new parsing logic in run_helpers_compound_tier.rs, integration across the run, debate, and review commands, and a refactor of csa-config to support compound parsing. Review feedback highlights the need to use ToolArg::from_str for more robust alias resolution, the importance of including the original tier name in error contexts for better diagnostics, and a recommendation to persist pre-execution errors in the review command to ensure consistency with other CLI tools.

[gemini-code-assist]

resolve_alias_to_tool uses parse_tool_name to resolve the canonical tool name from an alias. However, parse_tool_name only recognizes canonical tool strings (e.g., gemini-cli) and will fail for built-in aliases (e.g., gemini). This causes conflict detection to fail for user-defined aliases that point to built-in aliases. It should use ToolArg::from_str instead.

Suggested change

	fn resolve_alias_to_tool(alias_name: &str, cfg: &ProjectConfig) -> Option<ToolName> {
	let canonical = cfg.tool_aliases.get(alias_name)?;
	parse_tool_name(canonical).ok()
	}
	fn resolve_alias_to_tool(alias_name: &str, cfg: &ProjectConfig) -> Option<ToolName> {
	use std::str::FromStr;
	let canonical = cfg.tool_aliases.get(alias_name)?;
	match ToolArg::from_str(canonical).ok()? {
	ToolArg::Specific(t) => Some(t),
	_ => None,
	}
	}

[gemini-code-assist]

The error context for compound tier selector failures should include the tier name provided by the user to assist with diagnostics in the persisted pre-exec error result.

Suggested change

	tier_name: None,
	tier_name: ctx.cli_tier,

[gemini-code-assist]

Errors during effective tier resolution (including compound selector conflicts) should be persisted to the session directory to ensure consistent error reporting and observability, matching the pattern used in csa run and csa debate.

    let (effective_tier, args_tool) = match resolve_review_effective_tier(&args, config.as_ref()) {
        Ok(res) => res,
        Err(err) => {
            return Err(crate::session_guard::persist_pre_exec_error_result(
                crate::session_guard::PreExecErrorCtx {
                    project_root: &project_root,
                    session_id: review_pre_exec_session_id(&args),
                    description: Some(review_description.as_str()),
                    parent: None,
                    tool_name: explicit_review_tool(&args).map(|tool| tool.as_str()),
                    task_type: Some("review"),
                    tier_name: args.tier.as_deref(),
                    error: err,
                },
            ));
        }
    };