Flow behaviour:
1. In a node waiting for a button reply, how can we catch free text and
send the customer back to the AI node?
2. What triggers the AI node "request failed" port (timeout seconds,
retries, provider error codes)?
3. On a new Telegram chat, "thread started" and /start both fire, so the
welcome message is sent twice. What is the recommended setup?
4. Two quick customer messages make the AI reply take about 30 seconds.
Is there message batching, and can it be configured?

Google Sheets action:
1. Our Google Sheets integration stopped writing rows while still shown
as connected; reconnecting fixed it. Is there a failed-request log or
an alert?
2. Rows are written by column position and the header row is rewritten
with our keys. Can the action match existing headers by name?