Fixing matching:TestCheckIdleTaskList test flackiness

[dkrotx]

For some reason it almost always fails in IDE (4/5 cases, goland),
but passes in console. At the same time, the test clearly had issues:

• it's been 3 subtests in one
• only the first subtest checked task-list became stopped when unused
• the third subtest (the one which has been failing) never succeeded
  with tlm.AddTask since writer never started - instead of tlm.Start()
  we used special version which supposed to mimic it, but never started
  the writer.

The test is still problematic since it continues using time.Sleep all
around, but so do all the other tests for matching :-(

More comments and error-messages also bring more clarity to what the
test is actually for.

fix test flackiness in IDE

To be able to run tests in GoLand

make test + running all tests in GoLand

No risks, it is only the test been improved

Release notes

Documentation Changes

[davidporter-id-au]

sorry, small brain following along here, are we ensuring these daemons are started in the new test?

I 100% agree these tests are flakey and need some refactoring, but I'd also hate for us to lose testing of functionality, so it might just take me a minute to ensure they're equivalent

[dkrotx]

I looked at the code - these daemons are starting among with tlm.taskReader.Start():

for g := range tr.taskBuffers {
	go tr.dispatchBufferedTasks(g)
}
go tr.getTasksPump()

I think the previous approach where we started selected things bypassing the tlm methods actually a bad approach. We were trying to avoid notifyEvent (which is, I believe, no longer happening in .Start), but also missed some important changes like starting the writer.

I think it is really important for us to cover the happy-paths as well. Otherwise tests like TestAddTaskStandby pass unintentionally: it assumed we had to stop the writer though it's never been started!