fix: Update Groq vision models for PDF extraction

fix: update vision model to active qwen/qwen3.8-27b

fix: migrate from deprecated Groq Llama models to openai/gpt-oss-120b and 20b

Fix: actualizar modelo vision a qwen/qwen3.6-27b (los anteriores estan decomisionados)

Fix PDF extraction: logging, lower threshold, vision model fallback

Fix visuales y responsive para mobile