Dump cursors are currently tightly bound with file cursors and table cursors.

We should separate out the dump code so it is a separate layer, and the file and table cursors are oblivious to dumps. The dump cursor should configure the underlying cursor in raw mode and convert keys and values between the dump format and raw format.

This will make it easier to support dumping different cursor types and to implement different dump formats (e.g., CSV, XML).