feat: Route OpenRouter requests via Nebius with FP8 quantization

- Added OPENROUTER_PROVIDER_PREFERENCES config
- Prioritizes Nebius provider for better availability/speed
- Uses FP8 quantization for quality/speed balance
- Falls back to other providers if Nebius unavailable

src/lib/openrouter.ts

@@ -4,7 +4,7 @@ import OpenAI from 'openai';
  * AI Provider configuration
  * 
  * Set AI_PROVIDER in .env.local to switch:
- * - "openrouter" (default) - Uses OpenRouter with Gemma 3 27B
+ * - "openrouter" (default) - Uses OpenRouter with Gemma 3 27B via Nebius/FP8
  * - "gemini" - Uses Google Gemini 2.5 Flash
  */
 export type AIProvider = 'openrouter' | 'gemini';
@@ -38,3 +38,14 @@ export function getOpenRouterClient(): OpenAI {
 
 // Default OpenRouter model for vision tasks
 export const OPENROUTER_VISION_MODEL = 'google/gemma-3-27b-it:free';
+
+/**
+ * OpenRouter provider preferences
+ * - Prioritize Nebius provider for better availability
+ * - Request FP8 quantization for quality/speed balance
+ */
+export const OPENROUTER_PROVIDER_PREFERENCES = {
+    order: ['Nebius'],           // Prioritize Nebius
+    quantizations: ['fp8'],      // Use FP8 for quality
+    allow_fallbacks: true,       // Allow fallback to other providers
+};