Use the publisher's pinned SciFact object version for acquisition

[CompleteDotTech]

SciFact acquisition depended on an unversioned latest object. A clean current checkout now prefers the original publisher's immutable S3 object version, verifies its existing byte count and SHA-256 before installation, and falls back to the original URL only when necessary. --scifact-archive PATH supplies the same checked bytes locally when upstream access is unavailable.

The additive python -B -m graph_synthesis.datasets entry point prepopulates the ignored SciFact archive and then invokes the frozen downloader to prepare and verify all six original files. The graph workflow and graph extension guide use the current entry point. All 161 frozen files, expected digests, preparation logic, and dataset notices remain unchanged. No dataset is added to Git.

Source: the publisher's version 8LiW3OUBLBvhehDzrRdGD1ibIaXL.6PQ returned exactly 3,115,079 bytes with SHA-256 11c621288d41ac144d29b13b0f8503b3820b7d6e8b1f6ff24dff335c196d76be. Source provenance, versioning limitations, attribution, and recovery instructions are in graph_synthesis/DATASETS.md. The publisher can still remove that version or revoke access; this is a content-stable source, not a perpetual-availability guarantee.

Validation:

• 10 focused acquisition tests cover an empty cache, unavailable preferred source, valid fallback, changed responses, complete failure, local recovery, corrupt caches, failed writes, and offline check delegation.
• Windows Python 3.12 and a clean native Linux Python 3.12 clone each acquired all six original files from an empty data directory and passed scripts/download_datasets.py --check.
• Full portable offline replay passed on both platforms: 161 original files unchanged, 3,405 recorded calls, 3,644 predictions, zero network attempts, and 136 original tests passing without skips.
• Windows replay artifacts were byte-exact. Linux reported one accepted 3.47e-18 log-loss rounding difference under the existing unchanged cross-runtime tolerance.
• Signed commits and git diff --check verified.

Fixes #39. Related: #35 updates the main current-repository guide to the same entry point; #40 will retain this acquisition path when changing CI routing.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-28T18:34:01.887996Z	2342456	New commits
🔒 Security Review	✅ Completed 2026-09-28T18:13:18.252906Z	ac8fe64	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.