Measure the gas costs of processScheduledPieceMetadataRemovals for different numbers of deletions

[rvagg]

We do processScheduledPieceMetadataRemovals in nextProvingPeriod, what's the limit to that and can you schedule enough deletions in a proving period to prevent nextProvingPeriod from being called? We should know the limit and protect against this because it could make it impossible to progress with proving a data set.

[BigLep]

@rvagg : I have marked this as post GA (M4.5). Let me know if it needs to be done sooner.

[rvagg]

https://github.com/FilOzone/pdp/blob/323467e7b54a267d1774457d7b3321d94c6604ac/src/PDPVerifier.sol#L52

We should be limited to 2000 deletions during a period so maybe we're safe here. We should perform a test on this - queue up 2000 deletions in a proving period then see what the gas cost is of nextProvingPeriod and if it succeeds. That would probably tell us enough.

[rjan90]

This is now being actively investigated in FilOzone/pdp#283. The observed mainnet failures confirm the liveness risk described here: queued removals can make nextProvingPeriod unexecutable. This ticket remains useful for measuring the filecoin-services listener gas contribution and validating the eventual PDP fix.

FilOzone/pdp#286 is related because PDP state growth increases removal gas costs, but it does not replace the queue-liveness fix tracked in #283.