Run 30 paired trials of the baseline and candidate sorts

Run 30 paired trials of the baseline and candidate sorts

Alternate the baseline and the candidate over the same frozen inputs for 30 pairs, recording joules and wall time per run. Verify sorted output on every run and discard nothing silently. Deliverable: A CSV with trial, variant, joules, seconds, and output_hash, plus a log naming every discarded trial and its reason.