Improve performance of metricbeat elasticsearch module

**Describe the enhancement:**

During a serverless scale test, some elastic agent containers consumed more memory than expected, and a lot of this additional consumption was caused by the metricbeat elasticsearch module. The Nodes in question had in the order of ~50 Elasticsearch Pods on them, which is definitely an edge case, but @miltonhultgren indicated this could likely be significantly improved. I'm opening this issue to track those potential improvements separately from changes to agent itself. See https://github.com/elastic/elastic-agent/issues/5991 for more context.

If nothing else, it would be good to have a synthetic benchmark for this module.