Validate benchmark result shapes

[fallintoplace]

What changed

• Fixed the Snowflake L result row that had four values: [1.511,1.385,1,440] is now [1.511,1.385,1.440].
• Added a missing null row to the active Pandas c6a.metal result so it has the expected 43 query rows.
• Added validate-results.py for the main ClickBench result layout. Active dashboard entries now fail validation if their result matrix is not 43 queries x 3 runs, or if timings are not numbers/null. Historical and inactive shape issues are reported as warnings.
• Added a GitHub Actions workflow to run the validator on pull requests and pushes to main.

Why

Plain JSON parsing accepts rows like [1.511,1.385,1,440], but the dashboard expects three timings per query. That can make hot-run comparisons pick a fake 1.000 second timing instead of the intended 1.440 second timing.

Validation

• python3 -B validate-results.py
• python3 -B validate-results.py /Users/hoangvu/Code/OSS/ClickBench
• jq -e '(.result | length) == 43 and all(.result[]; type == "array" and length == 3 and all(.[]; . == null or type == "number"))' snowflake/results/20220701/l.json pandas/results/20260218/c6a.metal.json\n\nThe validator currently exits 0 and reports warnings for pre-existing historical or inactive files.