feat(evaluation): validate LongMemEval-V2 smoke inputs

[chenncy]

Related issue

Part of #1359.

Summary

• add a pinned LongMemEval-V2 dataset-lock and smoke-manifest validation contract
• verify the upstream harness checkout revision before writing smoke artifacts
• add a model-free smoke preflight command and usage documentation

Scope

This PR establishes reproducible smoke input validation only. It does not add the PowerContext Memory adapter, Reader/Judge execution, or benchmark scoring.

Validation

• python -m pytest -c evaluation/pyproject.toml evaluation/tests/unit/test_longmemeval_v2_catalog.py -q (10 passed)
• ruff check for the changed Python files
• git diff --check

AI usage

Implemented with OpenAI Codex under human direction and review.

[Copilot]

🟡 Changes recommended

The run manifest currently uses a misleading path_sha256 key for a content digest, which is a contract/naming issue that should be corrected before consumers rely on it.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Adds a reproducible, fail-closed validation layer for LongMemEval-V2 smoke inputs within the evaluation/ package, ensuring pinned dataset/harness identities and generating model-free “preflight” artifacts as groundwork for the broader benchmark work in #1359.

Changes:

• Introduces a pinned LongMemEval-V2 catalog + dataset-lock + smoke-manifest validation contract and harness checkout verification.
• Adds a model-free smoke-preflight implementation that writes non-overwritable manifest.json and subset.json.
• Exposes the workflow via powercontext-eval longmemeval-v2 smoke and documents usage.

File summaries

File	Description
evaluation/src/powercontext_eval/benchmarks/longmemeval_v2/catalog.py	Adds fail-closed loaders/validators for the pinned LongMemEval-V2 inputs, manifests, and harness revision.
evaluation/src/powercontext_eval/benchmarks/longmemeval_v2/smoke.py	Implements model-free smoke preflight artifact preparation and provenance manifest writing.
evaluation/src/powercontext_eval/benchmarks/longmemeval_v2/init.py	Introduces the LongMemEval-V2 benchmark package namespace.
evaluation/src/powercontext_eval/cli.py	Adds a longmemeval-v2 smoke CLI command for running the preflight.
evaluation/tests/unit/test_longmemeval_v2_catalog.py	Adds unit tests covering catalog validation, smoke selection constraints, digest pinning, and harness commit enforcement.
evaluation/tests/unit/test_cli.py	Adds a CLI test ensuring the new smoke preflight command can run without a model by patching the preparation step.
evaluation/README.md	Documents the LongMemEval-V2 smoke input validation workflow and required lock/manifest structure.

Review details

• Files reviewed: 7/7 changed files
• Comments generated: 1
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.