hermes - 💡(How to fix) Fix feat(telegram): streaming reply via sendMessageDraft (Bot API 9.5+)

Is your feature request related to a problem?

Yes. When using a slow model (e.g. o1, deepseek-r1), the user sees nothing for 10-30+ seconds before the full reply appears at once. The current streaming: true implementation uses editMessageText polling, which causes visible flickering and feels sluggish compared to native Telegram typing indicators.

Describe the solution you'd like

Telegram Bot API 9.5+ supports sendMessageDraft for all bots. This method streams partial message text to the user in real-time, appearing as a "typing" draft that updates smoothly — similar to how the native Telegram AI assistant works.

Proposed implementation:

1. On first token: call sendMessageDraft with partial text
2. On subsequent tokens: call sendMessageDraft again to update the draft
3. On completion: call sendMessage with the full message (replacing the draft)
4. Falls back to existing editMessageText if sendMessageDraft fails

This would provide a much smoother streaming experience on Telegram, especially for slow models.

Describe alternatives you've considered

Current editMessageText polling with edit_interval — works but causes visible flickering and re-rendering on each edit.

Additional context

• python-telegram-bot already supports sendMessageDraft since v22.x
• Prior PRs (#17179, #1312, #922, #774) attempted similar streaming approaches but were closed — this proposal specifically targets sendMessageDraft which is now available to all bots as of API 9.5
• Reference: Telegram Bot API - sendMessageDraft