KafkaJSNumberOfRetriesExceeded should reflect the original retry-ability

[JanGroot]

The KafkaJSNumberOfRetriesExceed now reflects if the original error was retryable.
If so: The onCrash will restart the consumer (regarding shouldRestartOnFailure)
If not: The onCrash will not.

closes #1271

[Nevon]

The intention is that the consumer should always retry if the reason for the crash was due to some temporary error that we could recover from by tossing away the current state and restarting. That's why we restart if the error is either retriable or KafkaJSNumberOfRetriesExceeded.

So the intention of this change, to make it so that we restart only if the error that caused us to run out of retries was a retriable one, makes sense. It's just unfortunate that the way to do it is to make KafkaJSNumberOfRetriesExceeded itself potentially retriable. At least to me, it seems unintuitive that an error saying that we're out of retries is retriable. It's kind of repurposing the property to mean something different than what it means for all other errors.

What do you think about keeping KafkaJSNumberOfRetriesExceeded as non-retriable, but change the restart check to e.retriable || (e.name === 'KafkaJSNumberOfRetriesExceeded' && e.originalError.retriable)?

[JanGroot]

Good point, above anything, it should be clear what the intention is. I have another idea, I'll update the PR.

[JanGroot]

New and more clear approach:
If the error was not retriable to begin with, reject with KafkaJSNonRetriableError.

closes #1271