More exact logging in storage and returning errors

[MaxiBoether]

I faced issues where for whatever reason in training the storage excited because we tried to access a sample > 0 in a jpg. I have not encountered the issue again, it might still be prevalent. This adds some more detailed logging on what is happening for further investigation, and also returns actual gRPC errors. I think the call propagation up the chain might also not be working as expected currently, but I don't have a log to test this with. Anyways, if you run into problems, having these logs helps, so let's merge this into main.

Step 4/n of merging SIGMOD changes.

[codecov[bot]]

Codecov Report

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

Project coverage is 82.51%. Comparing base (988bd01) to head (c528fce). Report is 3 commits behind head on main.

:exclamation: Current head c528fce differs from pull request most recent head 573c199

Please upload reports for the commit 573c199 to get more accurate results.

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #519 +/- ## ========================================== + Coverage 82.50% 82.51% +0.01% ========================================== Files 214 214 Lines 9955 9963 +8 ========================================== + Hits 8213 8221 +8 Misses 1742 1742 ```

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

[github-actions[bot]]

^{( % to main) ( % to main)}