Pass replied-to message text to LLM

[s-nagaev]

Goal

When a user talks to the bot via the reply feature, the text of the replied-to message must be included in the prompt that is sent to the LLM.
Right now only the user’s own message reaches the provider, so important context is lost.

Motivation

• Preserves conversational context (especially in busy group chats).
• Aligns behaviour with Hiroshi implementation (see s-nagaev/hiroshi#24).

Tasks

1. Update message-building logic
   • Detect message.reply_to_message in the incoming Telegram update.
   • Concatenate the original text and the user’s reply into one prompt, e.g.:

   > {replied_to_message.text}
   
   {user_message.text}
   

2. Provider-agnostic
   • Ensure the combined prompt is passed unchanged to any LLM provider (OpenAI, BigModel, etc.).

Reference

Similar fix in Hiroshi: s-nagaev/hiroshi#24