Implement context manager for usage metrics

[Nayjest]

metrics:

• total/new tokens: int
• total/new characters: int
• total/avg duration: float (seconds)
• total llm requests: int
• avg speed (tokens/sec, characters/sec) location: separate submodule (mc.metrics?) q: async? q: store distinct mertic objects in llm response object instead of gen_time + some other attributes? q: configuration options?