ci: add gallery compatibility workflow across OSes

[Copilot]

con-duct commands (ls, plot, etc.) must not fail when run against real-world log examples in con/duct-gallery. There was no CI enforcing this.

New workflow: .github/workflows/gallery.yaml

• Matrix test (ubuntu-latest + macos-latest): checks out con/duct-gallery, installs con-duct[all] from the current branch, then runs:

  • con-duct ls duct-gallery/logs/*/example_output_info.json
  • con-duct plot <info_file> --output <name>.png for every gallery example
  • Either command failing fails CI

• PR comment (non-fork PRs only): after the matrix jobs pass, pushes generated PNGs to an orphan gallery-pr-{N} branch and posts (or updates) a single PR comment with all plots folded in a <details> block — no comment spam on re-runs

• Triggers: pull_request, push to main, daily schedule (0 6 * * *)

• Permissions: minimal — contents: read for test jobs; contents: write + pull-requests: write only for the comment job

[yarikoptic]

@copilot the workflow run already failed. Can you experiment in a fork to which you have full access and provide us a solution which you would working?

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.00%. Comparing base (cc80272) to head (d828183).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #444      +/-   ##
==========================================
- Coverage   91.08%   91.00%   -0.09%     
==========================================
  Files          15       15              
  Lines        1245     1245              
  Branches      170      170              
==========================================
- Hits         1134     1133       -1     
- Misses         77       78       +1     
  Partials       34       34              

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

@copilot the workflow run already failed. Can you experiment in a fork to which you have full access and provide us a solution which you would working?

The root cause was that con-duct plot <info.json> failed when gallery logs are renamed: info.json says output_paths.usage = ".duct/runusage.json" but the gallery stores it as example_output_usage.json. The plot code took the basename (runusage.json) and looked for it as a sibling—but it didn't exist under that name.

Fixed in src/con_duct/plot.py (commit 3e2c892): when the referenced usage file isn't present next to the info.json, the code now falls back to searching the same directory for any file ending with a recognised usage suffix (*usage.jsonl, *usage.json). A new test (test_matplotlib_plot_info_json_renamed_usage) covers this case.

The ls command was already working fine—the first CI run showed it passing successfully.

[yarikoptic]

@copilot remind why you added this? there should be no magical "use some other file" logic IMHO

[Copilot]

Agreed — that fallback was an over-correction. Removed in commit revert: remove fallback usage-file logic from plot.py. The companion test has been removed as well.

The real fix needed here is in the gallery CI workflow (or in the gallery itself): con-duct plot should be pointed at the actual usage file (e.g. example_output_usage.json), not the info JSON, when the usage file has been renamed — so no "magic" is needed in plot.py.