Infinite Chocolate Fix

[jrhea]

This PR is a followup to PR #594 and provides a better fix to issue #587 in that it allows for infinite netting within a checkpoint (the previous fix simply disregarded the shortDeposit when calculating exposure for openShort()).

Summary of the Problem and Fix:
Edge cases were found fuzz testing that caused longs to be insolvent in rare cases. What would happen is that after the shorts were closed in _applyCheckpoint() it would underflow when trying to close out the longs. This would happen bc fuzzing found cases were a short would have a larger deposit than a long's fixed rate despite the short being for less bonds than the long (see issue #587). I determined that this was caused by numerical inaccuracies in fixed point math pow(). The fix was to round the short's base deposit part of the exposure calc down by the magnitude of the numerical error. This ensures that when LP's withdraw, the exposure is high enough to ensure there is enough left in the share reserves to allow longs to close at maturity.

Notes:

• New PoolConfig variable called precisionThreshold was added to make the fix configurable.

• resolves Exposure Symmetry Tests #662 by adding checkpoint exposure check in IntraCheckpointNettingTest.open_close_long_short

• In the process of implementing this fix, I audited OpenLong, CloseLong, OpenShort, and CloseShort for consistency.

OpenLong

• state deltas are calculated from result of deposit()
• _minimumTransactionAmount is checked against output from deposit

NOTE: This will cause some tests to fail that use minimumTransactionAmount as a lowerBound for fuzzing inputs. Including steth and sdai tests.

• minOutput is checked against values derived from deposit()
• bondProceeds is calculated from output of deposit and is checked against minOutput

CloseLong

• state deltas are calculated from the external function inputs
• minOutput is checked against output from withdraw()

OpenShort

• state deltas are calculated from the function input
• _maxDeposit shouldn't be checked against the output
  of deposit bc slippage could make a larger than maxDeposit
  potentially pass the check

CloseShort

• minOutput is checked against output from withdraw
• state deltas are calculated from the external function inputs

[coveralls]

coverage: 97.401% (-0.2%) from 97.592%
when pulling 6b2f220 on rounding-fix
into 2f638cb on main.

[github-actions]

Hyperdrive Gas Benchmark

Benchmark suite	Current: 6b2f220	Previous: 2f638cb	Deviation	Status
addLiquidity: min	755 gas	755 gas	0%	🟰
addLiquidity: avg	51475 gas	52287 gas	-1.5530%	✅
addLiquidity: max	97684 gas	97684 gas	0%	🟰
checkpoint: min	558 gas	558 gas	0%	🟰
checkpoint: avg	47612 gas	47626 gas	-0.0294%	✅
checkpoint: max	99085 gas	99410 gas	-0.3269%	✅
closeLong: min	755 gas	755 gas	0%	🟰
closeLong: avg	24357 gas	24257 gas	0.4123%	🚨
closeLong: max	114894 gas	114894 gas	0%	🟰
closeShort: min	713 gas	713 gas	0%	🟰
closeShort: avg	26958 gas	27198 gas	-0.8824%	✅
closeShort: max	113370 gas	113370 gas	0%	🟰
initialize: min	706 gas	706 gas	0%	🟰
initialize: avg	178066 gas	178066 gas	0%	🟰
initialize: max	252327 gas	252327 gas	0%	🟰
openLong: min	757 gas	757 gas	0%	🟰
openLong: avg	57852 gas	57649 gas	0.3521%	🚨
openLong: max	219289 gas	219128 gas	0.0735%	🚨
openShort: min	712 gas	712 gas	0%	🟰
openShort: avg	57104 gas	56978 gas	0.2211%	🚨
openShort: max	218776 gas	218660 gas	0.0531%	🚨
removeLiquidity: min	777 gas	777 gas	0%	🟰
removeLiquidity: avg	78634 gas	78286 gas	0.4445%	🚨
removeLiquidity: max	204374 gas	204374 gas	0%	🟰

This comment was automatically generated by workflow using github-action-benchmark.

[jalextowle]

LGTM