User should be able to initialize Kafka emitters eagerly when needed.

Currently emitters are initialized lazily until the first transaction is committed. If there is any issue, user won't be able to find it until emitters are initialized.
Because if there is a misconfig, there will be an issue/failure anyway. It would be better to let the problem surface as early as possible, instead of waiting till a transaction committed. This would save quite some time for debugging and testing.