fix: use mistralai SDK directly for embeddings — InferenceClient has no Mistral provider

fix: pass provider=mistral to InferenceClient — avoids Hub model lookup for mistral-embed

feat: initial enirtcod.fr MVP deployment