[sdk]: clear stale reverse mapping in setEidMapping

[royvardhan]

Summary

• HyperbridgeLzEndpoint.setEidMapping previously overwrote the forward map _eidToStateMachine[lzEid] and added a new reverse map entry, but never deleted the prior state machine's reverse entry. Inbound source authorization in onAccept reads only _stateMachineToEid[keccak256(request.source)], so a remapped or "disabled" EID continued to authorize messages from the previously-trusted state machine (incomplete revocation).
• setEidMapping now reads the prior state machine, deletes its reverse entry if non-empty, writes the new forward, and writes the new reverse only when stateMachineId is non-empty. Clean disable path with no keccak256("") residue.

Tests

Added fork-free unit tests at sdk/packages/lz-endpoint/test/HyperbridgeLzEndpointConfigTest.sol covering:

• testRemapClearsPriorReverseEntry — setEidMapping(101, A) → setEidMapping(101, B) leaves eidFor(A) == 0 and eidFor(B) == 101.
• testDisableClearsReverseAndAvoidsEmptyKeyResidue — setEidMapping(101, A) → setEidMapping(101, hex"") clears eidFor(A), does not populate eidFor(hex""), and isSupportedEid(101) is false.
• testSetEidMappingIsIdempotent — same args twice ends in byte-identical state.
• testReassignStateMachineAcrossEidsPointsReverseToLatest — admin remap of the same state machine across two EIDs leaves the reverse pointing to the latest EID; sanity check for behavior under admin mistake.

All 4 tests pass locally:

[PASS] testDisableClearsReverseAndAvoidsEmptyKeyResidue() (gas: 52399)
[PASS] testReassignStateMachineAcrossEidsPointsReverseToLatest() (gas: 90358)
[PASS] testRemapClearsPriorReverseEntry() (gas: 72670)
[PASS] testSetEidMappingIsIdempotent() (gas: 67532)

Split into a separate file from HyperbridgeLzEndpointTest.sol because the new tests are pure config-state checks and don't need the mainnet fork (MAINNET_FORK_URL) that the existing E2E tests require.