Describe the enhancement:

In some cases of high load, Elasticsearch will return 429 errors to indicate rate limiting.
Beats should back off if it detects a HTTP 429 response.

Looking at the code in here, it looks like it does not do that and just re-sends per code

// L500
if itemStatus == http.StatusTooManyRequests {
	stats.fails++
	stats.tooMany++
	return true
}

// L553
func (stats bulkResultStats) reportToObserver(ob outputs.Observer) {
	ob.AckedEvents(stats.acked)
	ob.RetryableErrors(stats.fails) // 👈 retries no back off
	ob.PermanentErrors(stats.nonIndexable)
	ob.DuplicateEvents(stats.duplicates)
	ob.DeadLetterEvents(stats.deadLetter)

	ob.ErrTooMany(stats.tooMany)
}

Describe a specific use case for the enhancement or feature:

I have an Elasticsearch cluster, and in my network I have deployed 1500 elastic-agents.
They are sending lots of logs to Elasticsearch, and I am routinely getting HTTP 429 errors.

On one hand, I am trying to scale the resources on the Elasticsearch server side.

However, it would be good if beats and elastic-agent could backoff if it detects HTTP 429 errors. Right now beats and elastic-agent seem to keep hammering on Elasticsearch if it returns HTTP 429 errors.