feat: add telamon.roundtable skill for parallel multi-agent discussions

[CasualCarlos]

Summary

• New telamon.roundtable skill that spawns 2-4 subagents in parallel for cross-cutting decisions
• Each agent thinks independently (separate Task call), producing genuinely diverse perspectives
• Orchestrator presents full unabridged responses — never synthesizes or blends
• Human drives follow-ups: "What does @architect think about @critic's point?"

How it works

1. Orchestrator identifies a question needing multiple perspectives
2. Picks 2-4 agents based on relevance (e.g. @architect + @po + @security)
3. Spawns all as parallel Task calls with context prompt (<400 words)
4. Presents each agent's full response separately
5. Human directs follow-up rounds

Inspired by BMAD-METHOD's "party mode" concept, adapted for telamon's real multi-agent architecture.

Files changed

• src/skills/workflow/roundtable/SKILL.md — the skill
• src/agents/telamon.md — skill reference