harvester_test: Fix flaky issues

- Fix "assert a harvester for same source cannot be started" test which was race-y, especially after #48445. The improved test correctly waits first for the harvester to be started and then starts multiple harvesters to assert they don't run despite the existing one.
- `WaitUntilIncreased` is unreliable and race-y. It can miss a goroutine starting before its invocation and it can have other edge cases as well, for example GC goroutines. I've completely deleted the function to avoid the foot gun.
- Fix multiple other synchronization failures discovered via the stresstest script.
- Some refactoring to use consistent timeouts, helper functions etc.
- Test "assert repeated start for same source does not leak goroutines with harvester_limit" is now merged with "assert a harvester for same source cannot be started"

Author: orestisfl