Handling of IO errors

[drahnr]

Currently any IO errors are going to cause a panic, for the most part since our trait design evolved around in-memory only. Now this is fine for the most part, most libraries like rocksdb do this internally, however, for custom code we require diligence to not i.e. panic on an EAGAIN which is non-fatal (not specific to SmtStore). These are two separate concerns, yet they're closely related - error handling of IO related errors.