feat: Migrate Prolific Worker Intelligence from A2A-SIN-Worker-Prolific to Bridge Server

[Delqhi]

Objective

Move all Prolific automation intelligence from the local Python worker to the Cloudflare Workers server.

What Moves to Server

• llm_decide() — LLM decision engine
• llm_verify() — Pre/Post-check self-reflection
• evaluate_study() — Study risk/reward scoring
• generate_persona() — Dynamic persona answers
• humanize_timing() — Human emulation delay calculations
• anti_detection_check() — Fingerprint analysis
• survey_answers.json — Profile data (encrypted in Supabase)
• Appeal logic — Rejection appeal generator
• Study bailout logic — Unsafe study detection

What Stays in Extension

• DOM extraction (already in content script)
• Action execution (click, type, scroll)
• WebSocket bridge to HF MCP (optional legacy path)
• Chrome extension popup UI

Migration Plan

1. Port Python llm_decide() to TypeScript in server/src/services/
2. Port survey_answers.json to Supabase encrypted storage
3. Update extension content script to call server API instead of local LLM
4. Remove direct OpenAI calls from extension
5. Test full flow: Extension -> Server -> Action -> Verify

Acceptance Criteria

• Zero business logic in extension
• All LLM calls happen server-side
• Prolific profile filling works end-to-end via server API
• Anti-detection logic is invisible to users