Multilingual inconsistency in docstring generation

Since the target language in the system prompt is not strictly specified, LLM probably tries to guess the language based on the context of a specific function. This results in bilingual files where some functions have Russian documentation strings (presumably if the code contains comments/output lines in a different language)