Fix table restoration for compacted changelogs

[wbarnha]

Fixes #176.

[wbarnha]

@elrik75 how about giving this a try? Apparently the offset - 1 logic was around since the beginning of the project and I changed a unit test to accommodate this.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 94.08%. Comparing base (a489db3) to head (45db5bb).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #373   +/-   ##
=======================================
  Coverage   94.08%   94.08%           
=======================================
  Files         102      102           
  Lines       11117    11117           
  Branches     1550     1550           
=======================================
  Hits        10459    10459           
  Misses        558      558           
  Partials      100      100           

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[wbarnha]

We'll need to test this further before merging.

[wbarnha]

I reviewed the code and the problem is much deeper than the changes here.

[elrik75]

This bug still exists. I recently lost my rocks-db disk and Faust started and launched the recovery. Once again some changelogs partitions did not start (they have an offset equal to the first Kafka offset - 1). Removing the -1 saved me again. I didn't see any issue after the "fix".

[wbarnha]

In that case, I'll reopen the PR as a reminder for me to investigate this.