Change use of block deadlines

[fhenneke]

This PR adds a grace period of one block to the deadline for settling on chain. It also removes the use of block deadlines in computing fees and other quantities.

The PR changes two things:

• Settlements are still rewarded if solvers settle one block after the deadline. This change is implemented to not punish solvers for slightly suboptimal transaction canceling in the driver. The alternative would be to implement very cautios canceling which might negatively impact revert rates.
• The use of block deadlines for setting surplus, fee, and execution_cost to zero was removed. For every successful settlemen, even if after the deadline by many blocks, those values will just be copies of the respective values in the settlement_observations table. The change to surplusand execution_costdoes not have any effect on solver payments. Those values are only used for analytics. The change to fee has an effect since it enters the computation of network_fee_eth. Before the change, the network fee was computed to be negative and essentially equal to -protocol_fee_eth. On the one hand, network fees were already removed from slippage using per order data. On the other hand, slippage is also modified by adding the value of network fees to reimburse solvers the cost of execution. Since network fees were computed to be negative protocol fees, instead of paying solvers those network fees they paid us protocol fees. See Potential bug with auctions being settled late #380. With the proposed change, network fees are always paid as part of slippage to the solver. This is consistent with how slippage is handled in general.

The estimated impact on solver from adding a grace period is an increase in overall rewards by about 1.48 ETH in the accounting period 2024-08-06--2024-08-13.
The impact of the second change on slippage is not estimated precisely. The overcharging of protocol fees amounts to 0.6 ETH. The missing reimbursement of network fees is not estimated.
See this Dune query.

The changes should be mirrored to dune-sync and data for the accounting period starting on 2024-08-13 should be re-synced.

[bram-vdberg]

LGTM!

[fhenneke]

Comparing the total performance rewards with this PR to the actual payouts for the last accounting period gives

Actual COW needed: 343097.8722
Actual ETH needed: 85.9442

New COW needed: 358344.1259
New ETH needed: 86.8461

Difference COW: 15246.2537 (around 1.19 ETH)
Difference ETH: 0.9019

The the estimates based on the Dune are somewhat accurate.