Add additional error handling for txn to increase stability

[ballard26]

Errors such as OperationNotAttempted and InvalidTxnState can be pretty common while doing transactions on clusters whose nodes are being started and stopped randomly. This PR attempts to recover a producer from these errors.

[jcsp]

Can you add error counters in the producer status struct?

I suspect there will be some transaction tests that expect no such errors, and some that will expect errors.

As I understand it, the transaction code should cope gracefully with intentional node shutdowns (where leaderships are transferred away first), but is allowed to hit an error when nodes unexpectedly stop. So each test case should know which kind of restarts it is doing, and be tolerant or intolerant of errors accordingly.

[ballard26]

Yeah, I'll add an error counter. Will talk to Bharath about whether this behavior is always expected or not. If not I'll add flag to enable this new behavior.