data(website): seed the new website category with 40,084 Wikidata records

[Seungpyo1007]

What

Seeds the website category added in TechEngine #48 with 40,084 records from Wikidata's Q35127 subtree, bumps the submodule to the engine that serves them, and publishes the dump.

Dataset: 1,113,452 → 1,153,536 records, and the API gains a 13th collection at /v1/websites.

Why this source

The dataset's own device categories are exhausted, so the question was what adds genuinely new entities rather than more of what we already hold. Two candidates were measured before committing to either:

• Steam / SteamSpy top-up of game — rejected. 90% of its titles already exist, and the apparent 9% remainder is mostly name variants of records we have (PUBG: BATTLEGROUNDS vs our playerunknowns-battlegrounds, Grand Theft Auto V Legacy vs grand-theft-auto-v). Importing it would have inflated the count with duplicate entities.
• Wikidata websites — 47,631 items in the subtree, no overlap with any existing category, and 89% carry an official URL. That is far better populated than the electronics categories probed earlier this session (digital cameras: 34 items total).

What was imported

47,631 QIDs → 40,273 English-labelled items → 40,084 records. 189 dropped for an unusable name, 243 slug collisions took a -<qid> suffix, and Windows-reserved stems (CON, PRN, LPT1…) are skipped — a con.json from the game import previously made the tree unindexable by git.

Fields are written only where the source had them: homepage_url (85% of imported rows), launch_date, languages, owners. Everything is verified: false and cites its Wikidata item.

Notes on the dump

• The submodule moves to the engine commit carrying both the category and the seed fix from TechEngine data(soc): GPU enrichment round 2 (13 older chips) #49 — without that fix, seeding this tree failed on the first insert (SQLite Date type only accepts Python date objects as input) because the loader coerced dates by the field name release_date and this category's column is launch_date.
• A partial dump writes a manifest describing only the collection it generated, so the websites entry was merged into the real site/public/v1/index.json rather than replacing it — the other twelve collections are intact (verified in the diff).

python -m app.validate → Data validation passed (the bundled validator gained the matching WEBSITE_REQUIRED rule).

Closes #1