feat: add AdLint post-demo backlog workflows

[ftchvs]

Summary

• Adds the post-demo AdLint backlog as one cohesive workflow update for reviewer handoff, metadata-only creative asset QA, and false-positive discipline.
• Keeps privacy boundaries explicit: review packet manifests omit raw ad copy, landing-page URLs/HTML, raw creative paths, and raw media.
• Adds a strict near-miss eval target so benign or review-only copy cannot pass merely on decision accuracy if policy false positives appear.

Changes

• adlint review-packet builds a local handoff packet from batch archives or single report directories with manifest.json, review-packet.md, copied Markdown reports, and policy source links.
• Creative asset serialization now includes declared metadata validation for MIME type, dimensions, duration, file size, and asset-type compatibility without reading media.
• evals/run_eval.py reports false_positive_pressure for guardrail rows and supports --max-policy-false-positives.
• Adds evals/datasets/false_positive_near_miss_v1.jsonl and make false-positive-near-miss.
• Updates README, PRD, and eval docs for the new commands and boundaries.

Test plan

• make false-positive-near-miss
• make creative-assets-eval
• CLI review-packet smoke against examples/batch_campaigns.csv
• make pr-preflight
• git diff --check
• make test -> 217 passed
• Focused pytest for review-packet and invalid creative metadata after final cleanup

Notes

• Completes Linear AND-264, AND-262, and AND-263 pending review/merge.
• Raw media analysis, OCR, frame analysis, and speech-to-text remain explicitly out of scope.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9e0de09828

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Resolve relative batch report paths from the cwd first

When users follow the documented workflow with a relative batch output dir such as adlint batch ... --output-dir reports/batch, run_batch records row report paths like reports/batch/risk.md in the summary, not paths relative to the archive. This resolver then prepends the archive directory and looks for reports/batch/reports/batch/risk.md, so adlint review-packet reports/batch ... aborts with a missing Markdown report even though the file exists. Please handle relative paths already rooted at the invocation cwd/output dir before falling back to source / path.

Useful? React with 👍 / 👎.