kibana OOM; captureSpanStackTraces:true implicated

[trentm]

Repro details: elastic/kibana#118560

This required (gleaned from some private chat) having APM enabled in Kibana with captureSpanStackTraces: true. That was accidentally set to true because the serverUrl was customized plus some complexity in how @kbn/apm-config-loader works.

• Kibana: v7.15.1
• APM agent version: v3.21.1, but likely happens back to v3.16.0 (which is the min version specified in Kibana v7.15.1's deps).

Ideas:

• Straight-up leak; or
• A combination of: (a) capturing stack traces is so slow, and (b) capturing results in retaining references that indirectly keep the ES response bodies referenced while capturing, and (c) there were enough ES queries going on that it got backed up and eventually sucked all memory.