Extract error causes from backtraces

[unflxw]

Python error backtraces already contain the information that we'd need to populate the "error causes" sample data, but the agent/processor currently merges all the backtraces from the reported error and its causes together.

Modify how backtraces are processed so that only the backtrace from the reported error is present, and extract the name and message of the derived errors into the "error causes" sample data.

See appsignal/appsignal-ruby#1010 for the Ruby implementation of this same feature -- though note that, in the Python case, no changes need to be made to the Python integration itself.