Refactor archiver tests

[krivard]

Based on #1522

Description

Type of change (bug fix, new feature, etc), brief description, and motivation for these changes.

Archiver tests are a twisty maze of numbered csv files, each file testing a different corner case. The before, after, and expected .diff content is defined in different places in the code, sometimes multiple times. This approach is difficult to interpret without spending extensive time with the code, making it (1) difficult to update, (2) difficult to review, and (3) easy for errors or inconsistencies to creep in.

This PR refactors the archiver tests to:

• collect information about each file together
• only compute expected data once
• name files with the corner case they test, so it's easier to keep track

Changelog

Itemize code/test/documentation changes and files added/removed.

• test_archive.py: substantial refactor
  • Example class
  • Expecteds class
  • CSV_BEFORE, CSV_AFTER, various expected_csv[0-9]_diff definitions merged into a single CSVS dict
  • csv[0-9] converted to descriptive names eg "deleted" "added" "unchanged"
  • new global EXPECTEDS storing the filenames expected in various locations at different stages of archivediffing
  • repetitive manual lists of expected filenames converted in most cases to computational definitions based on the CSVS dict, with an assert check wherever changes to the test cases must be manually propagated to EXPECTEDS
  • ArchiveDifferTestlike class with centralized definitions of cache/export setup and final filtered exports check
  • File content spot-checks converted to exhaustive checks
  • TestArchiveDiffer, TestS3ArchiveDiffer, TestGitArchiveDiffer converted from CSV_BEFORE etc to CSVS and EXPECTEDS

Fixes

• Working with tests should not feel like this:
  

[dshemetov]

Thanks for doing this, definitely overdue! Few minor suggestions, but overall looks great 🚀

[krivard]

this was excellent, thank you