feat: chat history memory — multi-turn AI conversations with LLM-powered search query refinement

- chatHistory array tracks user/assistant messages (10-turn cap)
- LLM-powered refineSearchQuery() rewrites follow-up queries via M.requestAiTask()
- fallback proper-noun extraction when model is unavailable
- all 5 workers inject history between system prompt and user message
- Clear Chat resets conversation memory
- consolidated worker files to public/ (deleted 6 root-level duplicates)
- fixed worker shadowing bug where public/ copies overrode root source files

Author: ijbo

README.md

@@ -459,6 +459,7 @@ TextAgent has undergone significant evolution since its inception. What started
 
 | Date | Commits | Feature / Update |
 |------|---------|-----------------|
+| **2026-03-13** | | 💬 **Chat History Memory** — AI chat now maintains conversation context across turns; `chatHistory` array tracks user/assistant messages (10-turn cap); LLM-powered `refineSearchQuery()` rewrites follow-up questions into self-contained web search queries via `M.requestAiTask()`; fallback proper-noun extraction when model is busy; all 5 workers (Qwen, Groq, OpenRouter, Gemini, LFM) inject history between system prompt and user message; "Clear Chat" resets memory; consolidated worker files to `public/` (deleted 6 root-level duplicates that shadowed source in Vite dev) |
 | **2026-03-13** | | 🔍 **4 New Search Providers** — added Tavily (AI-optimized, returns clean summarized results for LLM injection, 1,000/mo free), Google Custom Search Engine (official Google results, 100/day free), Wikipedia API (free encyclopedia search), and Wikidata API (free structured knowledge); total providers now 7; updated all dropdown selectors (AI panel, AI cards, Agent cards); Wikipedia/Wikidata skip API key prompt |
 | **2026-03-13** | | 🔍 **Search thinking block** — web search results now appear in a collapsible "thinking" `<details>` block *before* the AI response streams; two-phase UX: spinner appears instantly when search starts, populates with results (or "no results") when complete; source citation pills below; removed duplicate inline search details from AI response bubbles; fixed duplicate user message bug in `sendToAi()` dedup check |
 | **2026-03-13** | | 📐 **LaTeX coding block** — new `📐 LaTeX` toolbar button in Coding dropdown inserts `$$...$$` display math blocks; default template `\frac{\sqrt{2025} + \sqrt{3025}}{\sqrt{25}}` evaluates to 20 via Nerdamer; MathJax renders in preview; Playwright test added |