docs(benchmark): 落地 OpenPI 54-cell ARM64 诊断报告

[627150795]

Problem

Issue #46 已经记录了 54-cell ARM64 source-build-derived 诊断结果，但原 PR 将报告放在 docs/research/，缺少 Benchmark 类别所需的治理元数据和可检索的证据边界。

Value

把公开结果整理为可版本化、可引用的 Benchmark 记录，明确冻结条件、结果、限制和证据状态；同时明确这不是官方 Terminal-Bench 成绩，也不把小样本观察写成总体排名。

Approach

• 将报告移至 docs/benchmarks/OPENPI_PI_OMP_54_CELL_DIAGNOSTIC.md。
• 增加 draft 状态、创建/复核日期、适用范围、模型与 thinking level、任务、样本量、验证器、隔离条件、失败分类、核算方式和复跑边界。
• 新增 docs/benchmarks/receipts/openpi-issue-46-arm64-54-cell-v1.md，记录 18 条三次重复结果明细及六组可重算汇总。
• 将 Issue 和 release 链接统一为 openpi-dev/openpi 的官方地址。
• 不提交原始 JSONL、Sessions、日志、candidate workspace、缓存或凭据；收据保持 draft，直到有独立复核入口。
• 本 PR 只修改 Benchmark 文档和无凭据收据，不修改 OpenPI 运行时代码。

Validation

• 已基于当前 main（a15a0d88837f580187cbb02d7fcab91bff8f0d34）完成 rebase。
• git diff --check 通过。
• 收据一致性校验通过：18 条明细、六组 pass/fail/indeterminate 汇总全部一致。
• 已核对 Issue [Benchmark] OpenPI 0.3.0 / Pi / OMP：GLM-5.3 与 GPT-5.6 Luna 54-cell ARM64 派生诊断 #46 的公开结果、冻结版本、任务 revision、调度 hash、lock fingerprint 和安全边界。
• Windows 本地全量格式检查仍受仓库既有 CRLF 基线问题影响；本次没有修改运行时代码。

Impact

• 用户可见行为：新增一份位于 docs/benchmarks/ 的 Benchmark 报告及边界收据。
• 模型可见工具与运行时：无变化。
• 配置、artifact 和兼容性：无变化。
• 该记录仍是 draft，不宣称独立复算或官方 Benchmark 接受；原始证据包仍需由维护者提供可审阅入口。

Fixes #46

[tt-a1i]

This Benchmark report remains open under #277. Its merge gate is source-evidence review against Issue #46: frozen revisions, task/model/verifier identities, accounting, failure classification, limitations, and evidence references must all read back correctly. Documentation alone is not runtime or official Benchmark acceptance.

[tt-a1i]

Blocking evidence finding on exact head 8841a68d81bef7efe481aa57833af35201921808:

The prose matches Issue #46, and the underlying GLM/Luna evidence directories still exist locally, but this PR provides no retrievable evidence reference from which another reviewer can inspect or recompute the tables. Issue text, hashes, and a local-only archive are provenance pointers, not independently reviewable evidence. Under #278's Benchmark contract, keep this report unmerged until it is moved to docs/benchmarks/, given the governed metadata/status, and linked to a stable credential-free archive identity or bounded receipt that reviewers can retrieve. Do not bulk-publish raw Sessions, logs, credentials, or candidate workspaces.

Also replace the historical tt-a1i/openpi Issue/release links with the canonical openpi-dev/openpi identity (or explicitly label them as historical redirects).

No runtime finding; this is a publication/evidence-state gate.