feat: add BabylonSpoke with its own cap-bounded liquidation logic and per-reserve bypass flags

[avniculae]

Alternative to the feat/babylon-spoke branch. All Babylon behavior lives in a BabylonSpoke subclass and a dedicated BabylonLiquidationLogic library; the canonical LiquidationLogic is byte-identical to main and Spoke's only change is virtual on liquidationCall (no bytecode change for canonical instances, so gas snapshots stay at main).

BabylonLiquidationLogic mirrors the canonical sizing and reuses its building blocks (_validateLiquidationCall, _liquidateCollateral/_liquidateDebt, _calculateCollateralToLiquidate, bonus and target-HF formulas) instead of forking them. Its own param structs carry a LiquidationOverrides input, and sizing differs from canonical in three ways: the collateral available for seizure is min(suppliedShares, maxCollateralSharesToRemove) (the cap slots in where the canonical full-collateral bound sits), the dust threshold is a parameter (zero disables dust protection), and bypassTargetHealthFactor sizes the repayment to min(debtToCover, full reserve debt). When the cap binds below the user's balance, the remaining collateral and debt must respect the dust threshold or the call reverts with MustNotLeaveDust. A zero cap degenerates to a zero-amount liquidation, which the Hub rejects, so there is no bespoke zero check.

BabylonSpoke disables the canonical entry (UnsupportedLiquidationCall), exposes the overload with maxCollateralToRemove (uint256.max = no cap), and stores per-reserve LiquidationBypass flags (restricted updateLiquidationBypass; a flag on either reserve of a liquidation applies). The liquidation manager gate comes in a follow-up.

BabylonSpokeInstance exceeds EIP-170 at the standard spoke profile (24,952 bytes at via-ir/750), so it compiles under a new via-ir/200 profile (23,844 bytes, 732 margin). Slimming the contract instead is an option if the lower runs are a concern.

Tests: an equivalence test pins Babylon sizing with neutral overrides to the canonical output; deterministic unit tests cover cap/bypass/dust-threshold sizing including the dust-bump-exceeds-cap revert; an integration suite upgrades a spoke proxy to BabylonSpokeInstance in place. No canonical test or snapshot is modified.

Note for reviewers: repeated fuzz runs on this branch surfaced a pre-existing failure in test_getUserRiskPremium_fuzz_two_users_two_reserves_borrowed that reproduces on unmodified main (forge test --fuzz-seed 0x640, fails within ~20 runs with HealthFactorBelowThreshold). Unrelated to this change; happy to file it separately.

[avniculae]

Superseded by #1345: we're going with the dedicated Babylon liquidation logic scoped to the vaultBTC use case instead of generalizing the canonical LiquidationLogic. Keeping this branch around as a reference for the generic-overrides approach.