Add Retry-on-Failure Option for API Steps (OpenAI, Claude, etc.)

Currently, when an API call fails due to transient network issues — like timeouts or incomplete chunked reads — the flow fails entirely without retrying. This is especially problematic for long prompts or complex chains, where a single temporary glitch can break the entire automation.

Suggested Feature:
Add a "Retry on Failure" toggle to agent steps that involve external API calls, especially AI models. This should include:

• Number of retry attempts

• Delay between retries (ideally exponential backoff)

• Optional retry only on specific error types (e.g., APIConnectionError, 5xx errors)

Benefits:

• Greatly improves automation stability

• Prevents random AI response failures from breaking mission-critical flows

• Saves cost and time by avoiding full reruns

• Already standard in many automation platforms (e.g., Zapier, Make, n8n)