Tier 4: audit repository deposits behind DAS links #50

Follow-up to the #47 tier structure, adjacent to #35 and #38.

The corpus records where availability statements point (washdev carries das_repo_url; the datapapers backfill resolved repositories for the 8 data papers), but nobody has followed the links. We know 1 in 57 articles in the three journals without an enforced policy points to an open repository; we do not know how many of those pointers lead to a live deposit, in what formats, with what documentation.

Scope, mirroring Tier 3's committed-manifest pattern:

• Build the worklist: extract repository URLs/DOIs from DAS text and das_repo_url across washdev, ploswater, and the IWA snapshots (ws, jwh), one row per article-link with source journal and DAS excerpt
• Resolve each link: HTTP status (link rot is part of the finding), redirects, platform classification by domain (Zenodo, Figshare, Dryad, OSF, Dataverse, GBIF, NCBI/GenBank, DHS, IEEE DataPort, GitHub, institutional, other)
• Inventory deposits where platform APIs allow (Zenodo, Figshare, Dryad, OSF, Dataverse, GitHub): file names, formats, sizes, license, presence of documentation (README, data dictionary, codebook by filename heuristics)
• Classify per deposit: alive or dead; machine-readable share (csv/tsv/json vs xlsx vs docx/pdf); documented yes or no
• Committed outputs under data-raw/: the manifest CSV with per-link results (signature-stripped URLs, statuses, sha-free), a summary CSV, and a seeded stratified validation sample for human rating before the numbers enter the paper
• Caches of fetched metadata stay gitignored; requests paced and resumable; API-less platforms (Dryad tokens, bot checks) documented as gaps, following the Tier 3 precedent

Motivation for the paper: this turns 'a CSV without documentation is still closed data' from an argument into a measured finding, and completes the claim chain: statement, supplement (Tier 3), deposit (this tier).

Refs #47, #35, #38