Prompt refinement can now use live research

Research-aware refinement uses focused web sources when a request needs current or external information.

PromptDock now detects when a request would benefit from current information, documentation, comparisons, or a supplied URL, then uses Firecrawl as an optional server-side research layer before the existing prompt architect runs. Retrieved sources are bounded, deduplicated, preserved by URL, treated as untrusted data, and reused during clarification follow-ups. Ordinary creative and transformation prompts continue through the normal refinement path, while unavailable research falls back gracefully without pretending that sources were used.