[DNM] tests: backward-compatibility with v0.8.2

[akerouanton]

- What I did

We recently introduced bad-looking regressions in v0.9.0. These were not caught by our test suite. This commit adds a new backward_compatibility suite to make sure it doesn't happen again.

- How to verify it

New GHA steps should be green.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 43.04%. Comparing base (833d2c3) to head (a3251c4).

❗ There is a different number of reports uploaded between BASE (833d2c3) and HEAD (a3251c4). Click for more details.

HEAD has 13 uploads less than BASE

Flag	BASE (833d2c3)	HEAD (a3251c4)
	14	1

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #365      +/-   ##
==========================================
- Coverage   50.15%   43.04%   -7.12%     
==========================================
  Files          13        9       -4     
  Lines         650      460     -190     
==========================================
- Hits          326      198     -128     
+ Misses        279      233      -46     
+ Partials       45       29      -16     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[thaJeztah]

I think most (if not all) assertions we use a fairly basic; how much work would it be to implement these tests without introducing gotest.tools as dependency? (Mostly trying to keep the list of required dependencies as minimal as possible for this repository).

[akerouanton]

Tests are timing out on macOS due to the keychain being locked, and macOS displaying the GUI popup that requests password. I can't find a proper way to bypass that -- maybe security authorizationdb write … could do the job but the list of keys it accepts are undocumented and are subject to change across (major) macOS versions.

In retrospect, having tests that ensure backward compatibility is no different than a proper suite of conformance tests, plus human review ensuring tests aren't inadequately changed across versions. Of course, this means we can't ensure backward compatibility between version N (eg. v0.9.2) and version N-1 (eg. v0.8.2), but going forward it should be okay.

So, I'll abandon this PR and work on a shared suite of conformance tests instead.