[Reporting] Selectively partition async calls in generating CSV

[tsullivan]

#120309 changed CSV to asynchronously generate each row of content, rather than synchronously. This solved problems with CSV generation adding a single huge task to the Event Queue.

Problem: now that generating each row of CSV content is asynchronous, there is a log of context switching as there are as many async partitions as there are rows in the output CSV.

Solution: we can remediate the high amount of context switching by adding in async calls every N iterations of looping over each row.

We need to set up some performance tests to find what an ideal N value is.

[elasticmachine]

Pinging @elastic/kibana-reporting-services (Team:Reporting Services)

[elasticmachine]

Pinging @elastic/kibana-app-services (Team:AppServicesUx)

[tsullivan]

Closing this for now, since the impact seems low, and the cost of measuring the improvement seems high. We can always re-open this later if it starts to make better sense to do this.