Progressively calculate /_disk_usage?run_expensive_tasks

[flash1293]

Description

/_disk_usage?run_expensive_tasks is very helpful for admins to spot wasteful configurations. But it's not really feasible to run on production systems.

By obtaining information about field/data-structure level disk usage when the data is anyway moved around (e.g. during segment merge) and storing it in aggregated form as index meta data, fetching disk usage could get super cheap.

This feature could be exposed in the Kibana streams UI which is already showing approximate disk usage over time.

cc @dakrone @martijnvg as discussed

[pxsalehi]

This seemed like storage engine related. Although didn't find a matching label.

[elasticsearchmachine]

Pinging @elastic/es-storage-engine (Team:StorageEngine)