docs: optimize red teaming docs for MCP and humans

[Aryansharma28]

Summary

• Use attacker.marathon_script() instance method (auto early exit + backtrack padding) instead of standalone
• Add Check functions section with full ScenarioState/ScenarioExecutionStateLike method reference
• Add Exports section with exact Python/TypeScript imports
• Remove closed #2144 from roadmap
• Tighter prose: less narrative, more structured reference
• Every code example verified against source signatures
• All param names match implementation (fast_refusal_detection, detectRefusals, etc.)

Follow-up to #276.
Part of langwatch/langwatch#1713

Test plan

• cd docs && pnpm dev builds without errors
• All code examples match actual API signatures

🤖 Generated with Claude Code

[github-actions]

Automated low-risk assessment

This PR was evaluated against the repository's Low-Risk Pull Requests procedure.

• Scope: Modify red-teaming documentation: restructure and tighten prose; add "Check functions" and "Exports" sections; change example usage to use attacker.marathon_script() and updated parameter names/values; remove roadmap item #2144; adjust example turn counts and CI guidance.
• Exclusions confirmed: no changes to auth, security settings, database schema, business-critical logic, or external integrations.
• Classification: low-risk-change under the documented policy.

The PR only edits a documentation file (docs/pages/advanced/red-teaming.mdx), updating prose and example snippets without changing source code, configuration, secrets, database schemas, auth, or external integrations. All changes are documentation-level (example code in docs, explanatory text, roadmap items), which fits the allowed low-risk categories.

This classification allows merging without manual review once all required CI checks are passing and branch protection rules are satisfied.