Improve buffer settings of ES Loaders to allow for higher throughput

[jbeemster]

With v0.16.0 a lot of the memory settings were tuned which yielded fairly good results:

1. Large: 100-200 RPS (even split between good and bad)
2. XLarge: ~800-900 RPS
3. XXLarge: ~3000 RPS

However the common factor (outside of being CPU bound for throughput) was that the ES Loader could not keep up with the load. Digging into it this does not seem to stem from ES Cluster as an upstream bottleneck but rather that we are inserting 1 event at a time into Elasticsearch.

https://github.com/snowplow/snowplow-mini/blob/master/provisioning/resources/configs/snowplow-es-loader-good.hocon#L38-L43
https://github.com/snowplow/snowplow-mini/blob/master/provisioning/resources/configs/snowplow-es-loader-bad.hocon#L38-L43

This was presumably done so that events land in ES as soon as they are sent but it means that we have fairly terrible throughput at the higher end.

[jbeemster]

cc/ @istreeter as you had some ideas in this area for a future version.