[Epic] TPCH/TPCC-style benchmark matrix: Q1-Q8 (single-node first)

[ShuhaoZhangTony]

Goal

Move benchmark coverage from one-off scheduler comparison to a TPCH/TPCC-style matrix with Q1~Q8 workloads.

Scope

• Phase 1: single-node only (this epic)
• Phase 2: multi-node extension (separate issues)

Why now

Current HF upload pipeline aggregates whatever exists under results/**/unified_results.jsonl. We only see scheduler_comparison because we only ran that workload.

Deliverables

• Q1~Q8 single-node benchmark runner
• Unified result artifacts for each Q
• One-click aggregation + HF upload flow for the full matrix
• Summary docs + reproducible command set

Out of scope

• Multi-node scheduling parity and scale-out tuning (tracked separately)

Done when

• Running one command produces Q1~Q8 single-node artifacts
• aggregate_for_hf picks up all Q1~Q8 records
• HF dataset contains Q1~Q8 records with stable schema

[ShuhaoZhangTony]

审计结论：Q1–Q8 单机 runner、统一 schema 和使用文档已由 #28 补齐软件侧缺口，但 Epic 仍依赖 #13 的真实网关绿灯和 #14 的 HF 实际发布证据，因此现在不能标记完成。

本 Issue 不应继续由 PI 作为占位执行人。我已移除占位 assignee 并标记 help wanted；待 #13、#14 均完成后再关闭 Epic。