This ticket is tracking a very broad problem but I'm creating it for the sake of tracking a set of related FIXME comments in our source code. We have a few known issues with the column store read logic that need to be addressed as part of this project:

• At the moment, if there is no insert list, we don't call __wt_txn_read as part of cursor iteration. In durable history, we may still have an on-page value and associated history store values so this is no longer valid.
• When we can't find a visible value, we default to returning the on-page value. This is how things used to work pre-durable history but this assumption that on-page values must be globally visible is no longer valid.