Indexer auto scaling doesn't have any 'health' checks of workers

That leads to a situation when a host freezes due to some system issue, while Indexer auto-scaler still thinks his pool is ok, and pending tasks list keeps growing.

Start by tracing the Indexer auto-scaler’s worker-pool health handling and how it observes pending tasks. Define how a frozen host should be detected and how the pool should respond, then verify that unhealthy workers no longer make the pool appear healthy.

Written by the indexing model from the issue text.