perf(staking): O(1) operator stake aggregate + VPM share-pool slashing

[tangletools]

Summary

Two production-readiness scalability fixes at the staking layer. Both replace unbounded loops with O(1) accounting — making stake reads cheap inside the bill hot path and making slashing affordable for operators with hundreds of delegators.

1. O(1) operator delegated-stake aggregate

_getOperatorDelegatedStakeForAsset(op, assetHash) was O(B) — it summed _rewardPools[op][h].totalAssets + Σ_bp _blueprintPools[op][bp][h].totalAssets by walking every blueprint the operator had joined. This call sits on hot paths:

• TWAP accrual hook, fired on every delegate / undelegate
• Subscription billing _accrueOperatorWeights, fired per (operator, asset) per bill

For an operator on N blueprints, every read was N+1 SLOADs. Cost compounded inside the multi-asset bill loop.

Fix: maintain _operatorDelegatedAggregate[op][assetHash] incrementally. Every pool-totalAssets mutation now pairs with _increaseDelegatedStake / _decreaseDelegatedStake. Reads are a single SLOAD.

Mutation sites wired:

• RewardsManager._updateAllModePool / _updateFixedModePools (4 sites)
• DelegationManagerLib._setDelegatorBlueprintPosition (2 sites, Fixed-mode rebalance)
• SlashingManager._slashAllModePool / _slashBlueprintPool (2 sites)

Invariant: aggregate == rewardPool.totalAssets + Σ blueprintPool.totalAssets. Asserted by a new foundry invariant test that fuzzes 30,720 calls of delegate/undelegate/slash/reward combinations — passes with zero reverts.

2. VPM share-pool delegator slashing

ValidatorPodManager._slash was O(D) — it walked _operatorDelegators[operator] to proportionally debit each delegator's delegations[d][op] balance and emit DelegatorSlashed per delegator. For an operator with hundreds of delegators slashing became gas-prohibitive; with thousands it eventually exceeds block gas limits.

Fix: ERC4626-style share-pool. Each operator owns a DelegationPool { totalAssets, totalShares }; delegators hold shares in _delegationShares[d][op]. Slash is now a single SSTORE on pool.totalAssets; every delegator's effective claim drops proportionally via share→asset conversion at read time.

Inflation-attack mitigation: VIRTUAL_SHARES / VIRTUAL_ASSETS offsets in the Math.mulDiv conversion paths (Solady/OpenZeppelin pattern) defeat the first-depositor attack.

External ABI preserved:

• delegations(d, op) selector unchanged; now returns shares * (totalAssets + VIRTUAL_ASSETS) / (totalShares + VIRTUAL_SHARES). Automatically reflects slashing.
• operatorDelegatedStake(op) returns pool.totalAssets.
• OperatorPoolSlashed(operator, slashAmount, newTotalAssets, totalShares) replaces the per-delegator emission loop. Indexers reconstruct per-delegator impact off-chain from share balances + event.

Storage layout

UUPS-safe:

• DelegationStorage.__gap decremented 44 → 43; new _operatorDelegatedAggregate mapping appended at end.
• ValidatorPodManager appended new state (_operatorDelegationPools, _delegationShares) at end; no existing slot reordered.

Behavioral preservation

• Same external selectors, same return types.
• delegations(d, op) now correctly reflects slash impact without per-delegator state writes — a behavioral improvement, not a regression. Pre-PR slash updated delegations[d][op] directly; post-PR slash drops totalAssets, and the next read computes the slashed amount via conversion. Round-trip equivalent.
• Same Errors.* revert paths.

Test plan

• Full regression: 1,454 / 1,454 tests pass (+4 new invariants over baseline 1,450)
• New invariant invariant_operatorDelegatedAggregateMatchesPools: 30,720 fuzz calls, 0 reverts
• New VPM invariants (pool totals match net deposits, slash is O(1) gas-bounded, ABI views match share conversions)
• CI green
• Deploy to base-sepolia and slash an operator with ≥100 delegators; verify gas is constant w.r.t. delegator count

Production readiness checklist

• No storage layout changes that would break existing deployments
• No external ABI changes
• No new audit-round tags or session identifiers in code
• Invariant tests cover the new accounting in adversarial fuzzing
• Inflation-attack mitigation via virtual offsets

Depends on / related

• Sister PR Payments rearchitecture: RFQ hardening + multi-asset bill weighting + facet split for EIP-170 #133 (RFQ hardening + multi-asset bill weighting + facet split for EIP-170) is independent and can merge in either order. This PR branches off main so no rebase needed.