[kukirin-ui]

Adds a --json flag to slm runs list. It prints only a JSON array, one object per run with the run ID and everything from its status.json, so uv run slm runs list --json | python3 -m json.tool works. It goes through the same --project filter as the table, and an empty result is just [].

Tests in tests/unit/test_runs.py parse the output with json.loads, check both runs are there and that --project filters. Also added a CHANGELOG line under Unreleased.

ruff check/format are clean and the new tests pass. Locally the only failures in make test are the abc_music grader tests, which need abc2midi installed.

Closes #13

[sandeeps85]

Thanks @kukirin-ui, and thanks for asking on #13 first and laying out your plan. That's exactly how we'd like issues picked up. Sorry nobody answered you in time; that's on us.

What I checked

• make lint, and make test with abc2midi installed: everything passes, including your three tests. The grader failures you saw were only the missing tool.
• On real runs: slm runs list --json --project shakespeare_char gives the 3 reference runs, each with its full status.json fields, and | python3 -m json.tool works.
• Setting run_id from the directory name is a nice touch. Current runs already write it into status.json, but this way the output never depends on that.

I'll add the --json example to the command list in CONTRIBUTING.md on our side. Approving; merging once CI is green.