Configure Inngest retries and handle terminal failures with NonRetriableError, RetryAfterError and onFailure

Documented steps

1. Understand the default: in addition to the initial attempt, Inngest retries up to 4 times (5 attempts total).
2. Override per function with `retries: n` in the config object, where n is 0–20. `retries: 0` disables retries entirely.
3. For errors that will never succeed (validation failures, 404s, malformed input), throw `new NonRetriableError("message", { cause })` from `inngest` — it stops all further retries of that step/function.
4. For rate-limited upstreams, throw `new RetryAfterError("rate limited", retryAfter)` where retryAfter is ms, an ms-string, or a Date — Inngest schedules the next attempt accordingly instead of using its default backoff. Ref: https://www.inngest.com/docs/reference/typescript/functions/errors
5. To recover in-band, catch `StepError` in the handler after a step exhausts its retries and run a fallback step (`StepError` is available from SDK 3.12.0+).
6. For out-of-band cleanup, add an `onFailure: async ({ error, event, step, runId }) => { ... }` handler to the function config; it runs as a separate function triggered by the `inngest/function.failed` system event.
7. Alternatively, register a global handler on the `inngest/function.failed` trigger to route every failure to alerting. Payload: `data.error.{message,name,stack}`, `data.event`, `data.function_id`, `data.run_id`. Ref: https://www.inngest.com/docs/reference/system-events/inngest-function-failed
8. Read `attempt` from the handler args to branch behaviour on the final attempt.