feat: collect market fees through the Doppler initializer

[paoloanzn]

Doppler holds a pool's trading fees in its initializer and distributes them by share. collectFees(PoolId) harvests the outstanding amount and releases only the caller's portion, to a beneficiary registered when the pool was created. Uniswap V4 is a singleton, so a pool is a PoolKey rather than a contract, and its id is the hash of that key.

IDopplerHookInitializer declares that surface. Each signature was checked against DopplerHookInitializer and RehypeDopplerHookInitializer on Base Mainnet, and the seven-element getState tuple was decoded against the live contract:

getState(address)          0x1bab58f5
getShares(bytes32,address) 0x5ebb58fb
collectFees(bytes32)       0x817db73b

RIKRoyaltySplitter.collectPoolFees now takes the asset, reads the pool key from the initializer recorded at launch, derives the pool id, and credits the measured balance delta. Taking the initializer from storage rather than from the caller keeps collection from being aimed elsewhere.

Beneficiaries are fixed when a pool is created and cannot be added afterwards, so RIKLauncher reads getShares(poolId, splitter) after create and reverts with SplitterNotBeneficiary when it is zero. It also rejects a native numeraire, since releases are ERC20 transfers and the splitter cannot hold native value.

The mock reproduces FeesManager rather than approximating it — cumulative fees, per-beneficiary high-water marks, release to the caller — so the tests exercise the semantics Doppler actually has. MockAirlock delegates to the initializer the way the real Airlock delegates to poolInitializer.initialize. MarketSystem.invariant.t.sol absorbs the two invariants the splitter-only campaign held, which is removed.

The README documents the calling convention: the InitData layout, the beneficiary rules storeBeneficiaries enforces (ascending, summing to WAD, airlock.owner() with at least 5%), and the note that Airlock.create returns the asset in its pool slot for a V4 initializer.

286 tests pass, slither reports zero findings, lint and fmt are clean, and the deep soak runs 262,144 calls per system invariant.