Improve retry mechanism

[vivekjhauipath]

Typical flow is this, whenever darwin is invoked in a channel

• First the query is used to find relevant info from vector DB
• then we get access token
• finally we hit LLM gateway
• and we send response back to slack

Slack has separate retries => But we should make it exponential if not already.

But rest have a common retry, which is unnecessary and pollutes logs. We should have retries at more granular level, in most cases, a retry again LLM gateway call will suffice.