Fix: In order to ensure the "correct" ordering, *all* threads need to wait for the...

Fix: In order to ensure the "correct" ordering, *all* threads need to wait for the previous threads to finish - not only the threads that will emit a file. If both, thread n and n+1, want to emit a file and thus n+1 has to wait for n, we additionally must allow thread n+2 to terminate before n and n+1, even if n+2 does *not* emit a file. That's because as soon as n+2 finishes, it unblocks all threads x with x < n+2, which includes n and n+1. If that happens and n+1 is still waiting for n, we effectively allow n+1 to emit its file *before* n. By blocking all threads x with x > n+1 until n+1 has finished (regardless of whether they want to emit a file or not) this problem is resolved. As long as we assume that most threads actually *do* emit a file, this commit shouldn't hurt the performance.