Three agents per question and Assessment Copilot

What changed

• A question can now grant the AlgoArena agent alongside Claude Code and Codex. The candidate gets a tab for each agent, and each one keeps working while another is on screen.
• Recruiting Copilot is now Assessment Copilot. When you ask it for a terminal-agent question without naming a tool, it now grants all three agents, so candidates can use the one they know.
• The systems-design library gained 13 company-tailored questions, each paired with a harder assessment version.