feat: use Google Gemini as default vision backend, make provider configurable

- Plugin options: baseURL, apiKey, model, providerID, visionModels
- Default: Google Gemini OpenAI-compatible endpoint, gemini-2.5-flash
- Support {env:VAR} templates and GEMINI_API_KEY fallback
- Legacy fallback to ludmila-ai provider block for existing setups
- vision.md subagent now uses google/gemini-2.5-flash

agents/vision.md

@@ -1,7 +1,7 @@
 ---
 description: Анализ изображений и скриншотов (OCR текста или разбор UI)
 mode: subagent
-model: ludmila-ai/gemini/gemini-2.5-flash
+model: google/gemini-2.5-flash
 temperature: 0.1
 permission:
   read: allow