First problem:
For processes with a subprocess that contains an event subprocesses, the activity instance tree contains an activity instance for EITHER the subprocess OR the event subprocess but never for both.

Second problem:
Which one exists appears to be more or less random on how executions are returned from the database. Evidence: Asserting the activity instance tree for these cases randomly fails on some databases (postgres on Jenkins)

Possible solutions:

• make the event subprocess a scope
• fix the activity instance algorithm such that it is at least consistent and does not behave unpredictable (solving problem 2)

Once fixed, make sure to re-add the assertions removed here: https://github.com/camunda/camunda-bpm-platform/commit/242735625c2674846985c4fd42821cb89af234bd