numDelayed publications should only account for unproven publications in CheckpointTracker

[ggonzalez94]

Now that we allow passing older publications to the proveTransition function we must ensure that we only pay fees for the unproven ones in the ProverMananger.

• We are already doing so for the total number of publications since we use the latest proven checkpoint as a reference to calculate it.
• But for delayed publications we only ensure that they are smaller than the total number of publications. An attacker could choose older, valid publications that include delayed ones, and since they pay a larger fee use that to drain the contract.

We need to find a way to ensure that we only pay for new unproven publications or add that logic to the verifier somehow.