This involves:
- Adding storage for storing unit test results. I think for Differential this is stored as JSON and loaded, and we might just use the same mechanism for consistency, or we could have a dedicated table that actually stores the data into columns properly.
- Implementing a build step that runs "arc unit" consistently across different platforms. D9113 actually addresses most of the cross-platform issues that were encountered with running a specific command, so most of that should be able to be adapted to the "arc unit" test case.