Build failures cause unscheduled tasks which cause no results to ever be written to bugs

[tomrittervg]

See https://bugzilla.mozilla.org/show_bug.cgi?id=2052735 for an example.

A build failure in https://treeherder.mozilla.org/jobs?repo=try&resultStatus=unscheduled%2Csuperseded%2Ctestfailed%2Cbusted%2Cexception%2Csuccess%2Cretry%2Cusercancel%2Crunning%2Cpending%2Crunnable&revision=c08e22c0574f left many jobs in an unscheduled state (e.g. test-macosx1470-64/opt-gtest-1proc). Eventually they expired, but that takes 28 days.

[tomrittervg]

https://bugzilla.mozilla.org/show_bug.cgi?id=2062356

[tomrittervg]

After discussion, we came up with two paths forward:

1. The immediate fix: just ignore unscheduled tasks. These are recently added, and ignoring them brings us back to the old behavior. The risk here is if we ping treeherder inbetween a task completing, and the dependent-upon-it-tasks being scheduled, Updatebot will think the job is done when it is not. Very rare.
2. The 'best' fix: create a marker task in the task group that runs on built-in/succeed and does nothing but depend on the whole task group. Instead of checking to see if there are any tasks that still need to finish - check just this marker task. For 'build fails, tests never got scheduled' scenario, the unscheduled tasks would hit the 24h expiry and then the task would eventually complete (successful or otherwise.)