Improve Stacktrace truncation for StackOverflowError

[adinauer]

Problem Statement

We've had reports of dropped events in the past and in response implemented a simple limit of 100 frames.

This limit only uses 100 frames from one side of the stacktrace, leading to potentially unhelpful information shown in our product.

Solution Brainstorm

In the next major, we could improve the truncation to:

• only trigger on StackOverflowError so we do not lose information on smaller stack traces
  • this way we don't butcher other stacktraces
• truncate from the middle, i.e. keep frames on both ends of the stacktrace, hopefully truncating only some of the recurring part but keeping the most important frames, see relay
• keep 500 frames from both ends of the stacktrace, i.e. 1000 total frames to be compatible with relay potentially bumping their limit

NOTE: there is a risk of running out of sync with relays implementation / truncation numbers

[linear]

JAVA-208