If the resource is already being destroyed, we don't want to hop to the
`destroy` label again when the `destroy` semaphore is incremented.
We check the current label to avoid hopping again. Sometimes we forget
to update the list of labels to check when we add new labels after a
destroy.
Instead of checking the labels separately, we can increase another
semaphore, like `destroying`, along with the `destroy` semaphore and
keep it incremented until the resource is destroyed.