Investigate moving sourcemapping to an enrich processor

[axw]

I would like us to investigate moving sourcemapping logic out of apm-server, and into an ingest node pipeline. This would enable us to fix #2724, and would likely also speed things up by doing everything in Elasticsearch, where the sourcemaps are stored.

In order to do this, I think we could use the Enrich processor to enrich ingested documents with the sourcemap, followed by a script processor which adjusts/enriches stacktrace fields, and finally removes the sourcemap field.

To use the Enrich processor, we would need to store a field which concatenates the properties we use to match them into one field: service name, service version, and file/URL path.

[axw]

There are some ramifications to this change that I had not previously considered, as there are several things that depend on the stacktrace:

• error grouping key calculation (see also Investigate switching error grouping key calculation to use something faster apm-data#146), including replacement (or simply removal) of the apm-server.rum.exclude_from_grouping config
• identification of RUM library frames
• error culprit identification