Complete HeatOps v2 benchmark evidence and documentation

[MichelleHan7]

Why

Core v2 was merged in #3 while benchmark work was paused. This follow-up completes the measured evidence and documentation without changing the scheduling engine.

Verified results

• 165 synthetic instances, 660 policy records: 582 OPTIMAL, 72 FEASIBLE, 5 UNKNOWN, 1 NOT_RUN.
• All 654 returned schedules passed independent validation.
• 100 jobs / 10 crews: 11 of 12 policy attempts returned feasible incumbents; none proved optimal within the 30-second solve cap.
• Heat-first mean reduction: 8.47% on 117 valid, nonzero-baseline pairs out of 150 main scenarios. Median 3.67%; added delay and exclusions are reported.
• Original Phoenix result preserved at 2.971485%.

Artifacts

Adds raw CSV, summary JSON, compressed full schedules, three charts, source/environment metadata, preserved serial partial results, profiling measurements, methodology, architecture, a complete Chinese implementation report and three evidence-backed resume bullets.

Validation and limits

Core CI passed on Python 3.11 and 3.12 (162 tests); final Ruff/format/compile checks passed. Final timings combine a checkpointed four-process run with four resumed instances; resume metadata is included. Synthetic results are not real-world deployment or medical outcomes. No deployment settings or main-branch writes are performed by this follow-up.

Final follow-up CI also passed on commit 86f7947e4f8736c57d7b1ab52d43bca2c1a0c677: https://github.com/MichelleHan7/HeatOps/actions/runs/37886627179 . The branch remains a draft for review.