Explore using more Elasticsearch response codes to pause or stop model indexer autoscaling

[marclop]

As a follow up from #9463, it was suggested we may want to explore more HTTP response codes from Elasticsearch as a way to limit or stop autoscaling actions from taking place (or downscaling actions to be triggered).

Some other status codes we may consider is looking for 499 (client timeouts?), 502 and 503. All these may signal that Elasticsearch isn't fully responsive and that autoscaling should be paused if a certain rate is exceeded.

We're not collecting any of these metrics already in the model indexer stats and they would incur a slight storage cost.