Avoid setting signer cache on failed EVM broadcast
There was a follow on commit in PR to fix an AVAX double spend about a year ago (!2629 (18ef9070)) - this was not commented in the PR, but IIRC there was discussion in the war room at the time to just return nil and let it reschedule. This was fine at that time, but the combination of insta-observe and the recent change to vault rescheduling are incompatible - it needs to now return that error, otherwise it's added to the auto observations, which downstream call OnObservedTxIn
where it is added to the signer cache. Effectively it fails broadcast, the auto observations prevent reschedule, and the outbound is never again attempted.