feat(apps): add GPU server fallback to all LLM-using apps

Configure all apps with gpu-llm.mana.how as fallback when MANA_LLM_URL
is not set. This ensures apps can use the GPU server's local LLM models
(Ollama gemma3, qwen2.5-coder) instead of cloud providers.

Apps updated:
- Chat: LLM fallback to GPU server
- Context: LLM fallback (replaces Azure OpenAI dependency)
- NutriPhi: LLM + Vision fallback (replaces Google Gemini for food analysis)
- Planta: LLM + Vision fallback (replaces Google Gemini for plant analysis)
- ManaDeck: LLM + Vision fallback for card generation
- Traces: LLM fallback for AI city guides

Vision model default: ollama/gemma3:12b (multimodal, runs on RTX 3090)
Added VISION_MODEL to .env.development

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

apps/context/apps/backend/src/app.module.ts

@@ -26,7 +26,7 @@ import { HttpExceptionFilter } from './common/http-exception.filter';
 		LlmModule.forRootAsync({
 			imports: [ConfigModule],
 			useFactory: (config: ConfigService) => ({
-				manaLlmUrl: config.get('MANA_LLM_URL'),
+				manaLlmUrl: config.get('MANA_LLM_URL') || 'https://gpu-llm.mana.how',
 				debug: config.get('NODE_ENV') === 'development',
 			}),
 			inject: [ConfigService],