Having a multi-instance task (e.g. service task) that is async (before or after), custom job retry cycles are not evaluated. This is a regression due to the additional multi-instance body scope we introduced in 7.3:

• the MI body activity is now async (i.e. isAsync returns true)
• the inner activity is not (i.e. isAsync returns false)
• the class org.camunda.bpm.engine.impl.bpmn.parser.FoxFailedJobParseListener only checks for the inner activity