[Feature] Add language-based model/URL selection for LLM embeddings

[Feature] Add language-based model/URL selection for LLM embeddings

Support language-specific embedding models via language_models config:
- Shorthand: language_models = { ru = "model-name" }
- Full config: language_models = { ru = { model, url, api_key } }

Uses get_displayed_text_part() for language detection.
Include language in cache key for proper separation.