fix: #2308 declare CCIP approve with no outputs and narrow the LayerZero rationale

[suisuss]

Follow-up to #2389, from an independent review of that PR. Four fixes, none of which change runtime behaviour for a conforming token.

1. protocols/abis/ccip-erc20.json had the same defect, one file over

approve declared a bool output. Both contracts bound to that ABI - ccipBridgeToken and ccipFeeToken - are userSpecifiedAddress: true, and USDT is a CCIP-supported bridge and fee token. So on the EOA path the adapter's preflight staticCall decodes USDT's empty return as a bool, throws BAD_DATA, and the approve fails before it is broadcast - byte for byte the failure #2389 fixed for LayerZero.

The coverage suite cannot catch this. CCIP_ERC20_ADDRESSES points at WETH/WBNB/WMATIC/WAVAX, which all return bool, and both approve actions are skipped in the fixture regardless. Guarded here with a unit test instead; verified it fails when the bool declaration is restored.

2. #2389's comment overstated the failure, in the same direction as the comment it replaced

It said the mismatch "is not inert on the write path". Only the EOA path decodes:

Route	Preflight staticCall?
EOA direct	yes - breaks
Safe	no - encodeFunctionData + estimateGas
Safe-role	no - same
Turnkey sponsored	no - encodeFunctionData only

The comment it corrected claimed the write path never decodes; its replacement implied it always does. Both wrong, opposite directions. This matters practically: the same node, ABI and token fails for an EOA connection and succeeds for a Safe one, so a repro that omits the signer mode proves nothing either way.

3. One assertion in #2389's test discriminated nothing

decodeFunctionResult("approve", bool32) not throwing holds under outputs: [bool] as well as outputs: [], so it documented rather than tested. Dropped.

Worth noting what this test still is not: a guard on the mechanism. Nothing exercises EvmChainAdapter.executeContractCall against a provider returning "0x". A stubbed-provider test there would cover every write ABI at once, including lib/contracts/abis/erc20.json, which still declares bool on approve/transfer/transferFrom. Both are left to the separately tracked ticket rather than widened onto a release branch.

4. docs/plugins/layerzero.md documented an output that never existed

The OFT Approve page listed result | bool | Result. buildOutputFieldsFromAction gates ABI outputs on action.type === "read", so a write action surfaces success, error, transactionHash, transactionLink. That row was wrong before this change and contradicted the shipped ABI after it.

Verification

• tests/unit/protocol-layerzero.test.ts (22) and tests/unit/protocol-chainlink.test.ts (30) pass; protocol-calldata (508) unaffected
• The new CCIP test fails when the bool declaration is restored, so it is a real guard
• biome check clean on all five changed files