fix(credit): always use safe prices after partial-liquidation withdrawals

[SashaMIT]

Problem

partiallyLiquidateCreditAccount always withdraws collateral (fee in underlying to treasury, seized token to the liquidator), then runs _fullCollateralCheck with a hardcoded useSafePrices: false.

The multicall path treats any withdrawCollateral as requiring safe prices: after the withdraw (and after any phantom-token adapter call) it unconditionally ORs USE_SAFE_PRICES_FLAG into the flags used for the final check (CreditFacadeV3._withdrawCollateral, return flags | REVERT_ON_FORBIDDEN_TOKENS_FLAG | USE_SAFE_PRICES_FLAG). That includes the case where a phantom withdrawer's adapter returns false for useSafePrices.

So the partial-liquidation path could leave an account that passed on a manipulated main feed after collateral left the account, while the equivalent multicall withdraw would have forced the stricter min(main, reserve) check (and zero for tokens with no reserve feed).

Related: #313 only forwards flags & USE_SAFE_PRICES_FLAG from the phantom/external call. That is incomplete for ordinary (non-phantom) seizes, and also incomplete when a phantom adapter returns false (the in-repo PhantomTokenWithdrawerMock does exactly that). This PR supersedes that approach.

Fix

Pass useSafePrices: true on the partial-liquidation collateral check. Withdrawals always happen on this path, so this matches multicall withdraw semantics.

Verification

forge test --match-test 'test_U_FA_16_partiallyLiquidateCreditAccount_works_as_expected|testU_FA_16A_partiallyLiqudiateCreditAccount_works_as_expected_with_phantom_token'

Both pass. Unit expectations updated from false to true on the fullCollateralCheck call.

Made with Cursor