priority 1
----------
- join non-deduplicated chunks
  - choose when and how to
- delta encode chunks
- read from repo

priority 2
----------
- use more the `Reader` API (which is analoguous to the `IOStream` in Java)