feat: OffRamp-derived CCV selection for live CCIP 2.0 fork routing

[magiodev-cll]

Context

Follow-up to #54. The refresh (#66) has landed on devrel-205/cl-local-v3, and this branch is now rebased onto the resulting merge commit, so the diff is only this work. It supersedes the fork-side draft magiodev-cll#3.

Why

#54's V2 routing decodes MessageV1 locally to derive the CCV list and to call executeSingleMessage(MessageV1, ...). Today's live lanes encode finality as bytes4 (chainlink-ccip #1898, 2026-04-08) while the pinned dependency still decodes uint16, so live V2 fork tests fail with InvalidDataLength(MESSAGE_FINAL_OFFSET).

This ports the approach we validated in production rehearsal: let the destination OffRamp select the CCVs (getCCVsForMessage(encodedMessage)) and execute the raw encoded message through the permissionless execute entrypoint. No local message decoding, so routing keeps working as the wire format evolves.

What

• V2VerificationMode.OFFRAMP_DERIVED (new, fork-only):
  • destination chain resolved from the CCIPMessageSent topics, no MessageV1 decode
  • required CCV list read from the destination OffRamp
  • permissionless execute(encodedMessage, ccvs, verifierResults, 0), candidates tried in reverse router order, processed only on success
• IOffRampExecuteV2 gains the bytes-based execute entrypoint (next to executeSingleMessage).
• Test helpers ported onto V3: getOffRampForLane, setLaneDefaultCCVs, IOffRampSourceConfigV2Fork, and the CCVNoOpVerifier test double (src/test/ccip/CCVNoOpVerifier.sol), so a lane's default CCV can be mocked with a no-op on a fork.
• New e2e: test/e2e/ccip/CCIPv2Fork.t.sol — Sepolia → Arbitrum Sepolia message + token transfer over the live 2.0 lane, with the no-op CCV as the single counterfactual (validation, quorum walk, token release/mint, receiver call all production code).

Evidence

• forge test --match-path "test/unit/**" — 42/42
• forge test --match-path "test/e2e/ccip/CCIPv2Fork.t.sol" (public RPCs) — 2/2
  • Note: the other live-lane V2 fork tests in test/e2e/ccip still fail on the pinned codec path (STRICT/SYNTHETIC_ONLY modes); they pass once the dependency bump lands or once switched to OFFRAMP_DERIVED.
• Downstream rehearsal consuming this work: smartcontractkit/utb-avalanche-bridge#52 (e2e 5/5; its pin needs bumping from the pre-rebase 917cd63 to 1a4c265).

[magiodev-cll]

superseded by #68