chore: establish clean v0.2 baseline

Include only the working reconstruction prototype with Ollama integration:
- double-wrapping fix (lib/llm/provider.js)
- explicit v0.2 JSON output schema (prompts/reconstruct-v0.2.md)
- Zod validation layer (lib/reconstruction/schema.js)
- shared core analysis path (lib/analysis.js)
- prompt versioning infrastructure (lib/reconstruction/prompt.js)
- provider abstraction
- functioning Ollama provider path
- updated API route with centralized analysis
- UI components displaying v0.2 data and validation errors
- .gitignore rules for generated evaluation artifacts

Exclude: evaluator experiments, diagnostic tests, debug scripts,
generated artifacts, comparison findings, test data tied to evaluator.

lib/llm/provider.js

@@ -93,11 +93,9 @@ async function detectChatSupport(baseUrl) {
 
 class OllamaLlmProvider {
   async generateReconstruction(scenario, modelName) {
-    const { buildPrompt } = await import("@/lib/reconstruction/prompt");
-    
-    let rawPrompt = buildPrompt(scenario);
-    // Stronger JSON hint since we can't use format:json on older Ollama
-    const prompt = rawPrompt + `\n\nReturn ONLY a valid JSON object starting with { and ending with }. Do NOT include any text before the opening brace or after the closing brace. Do NOT wrap in markdown backticks.`;
+    // scenario is ALREADY a fully-built prompt text (built by analyseScenario).
+    // Do NOT call buildPrompt() again — that would double-wrap the prompt.
+    const prompt = scenario;
 
     const baseUrl = process.env.OLLAMA_BASE_URL;
     if (!baseUrl) throw new Error("OLLAMA_BASE_URL is not set");