Fix: Add fields to frame if it does not already exist when grouping by multiple terms

[kevinwcyu]

What this PR does / why we need it:

Multiple group by terms would only have the final group by term processed and added to the frame when executing on the backend.

This PR updates the way the frame is built up when processing aggregations. If the frame does not have any fields yet, the propKeys are added as fields (this will include all of the terms except the final term), otherwise if the frame already has fields, it will use those existing fields.

Which issue(s) this PR fixes:

Fixes #391

Special notes for your reviewer:

Sending queries to the backend needs to be reenabled for this to work. Change the following condition in datasource.ts

      request.targets.every(
        (target) =>
          target.metrics?.every(
            (metric) =>
              metric.type === 'raw_data' ||
              metric.type === 'raw_document' ||
              (request.app === CoreApp.Explore && target.queryType === QueryType.Lucene)
          ) ||
          (request.app === CoreApp.Explore && target.queryType === QueryType.PPL) ||
          target.luceneQueryType === LuceneQueryType.Traces
      )

to the following

      request.targets.every(
        (target) =>
          !(target.queryType === QueryType.PPL && target.format === 'time_series' && request.app === CoreApp.Dashboard)
      )

We can reenable sending queries to the backend after the remaining backend migration related bugs are fixed.

Here's an example of a query you can test with. It just needs to have at least 2 group by terms.

[idastambuk]

Tested it, looks good!
The only thing is that sometimes Im getting a too many buckets error and it's not passed on to the user (all we get in the UI is unknown error. I created this to make sure we don't forget it.

[njvrzm]

Small nit, one minor comment and one question but: 👍🏻

[njvrzm]

Tiny nit: you can just write if len(queryResult.Frames) != 0 here - you'd only need the nil check if queryResult.Frames was a pointer.

[njvrzm]

Curious - did something in the sdk or elsewhere change, or did this code path never get executed before?

[kevinwcyu]

this bit of the code used to build the results the same way as the frontend flow (building a table), but was refactored to use data frames instead. that refactoring was done by following elasticsearch's pr (which had the same bug that this PR is addressing). so it probably wasn't being executed from that point onwards. but we didn't notice because the backend flow would've only been run for alerts or in explore and it would've only been an issue if someone was using multiple group by terms.