Add escalation check: research asks before upgrading to deep-research

[5uck1ess]

Summary

• Adds an escalation check to the research skill after the corroboration step
• If sources conflict, stakes are high, or confidence is low, asks the user before upgrading
• Never auto-escalates — always asks first to avoid burning tokens

Escalation triggers

1. 3+ CONTESTED claims (sources actively contradict)
2. High-stakes domain (security, architecture, compliance, financial)
3. User expressed uncertainty ("is this actually true", etc.)
4. Most claims UNCORROBORATED (>50% single-source)

Test plan

• Run research on a simple question — should NOT trigger escalation
• Run research on a controversial topic — should offer to escalate
• Verify user can decline and get standard synthesis