Docs: dev port, wuchale OpenRouter key, root CLAUDE.md, UX findings (split out of #1281)

[michielbdejong]

Polle's (@Polleps) doc edits from #1281, split out so the Forms PR (#1923) stays about forms. The commit is his.

• CONTRIBUTING.md: the data-browser dev server runs on 6747, not 5173 (server.port in browser/data-browser/vite.config.ts).
• browser/CONTRIBUTING.md: wuchale's AI translation needs OPENROUTER_API_KEY (see openRouterTranslator in wuchale.config.js), not a Gemini key.
• browser/README.md: the demo video link as an autolink.
• CLAUDE.md (new, one line: @AGENTS.md), so Claude Code reads the repo's agent guidance.
• planning/Findings 2026-07-22.md: UX findings from a test session.

Left out of #1281's doc edits:

• the AGENTS.md title change (# AGENTS.md → # ATOMIC DATA), which looked accidental;
• the index-performance.md update, which documents code in Forms (#875), continuing Polle's #1281 #1923 and moves there.

🤖 Generated with Claude Code

[michielbdejong]

CI status (atomic-server worker, taking over from the Forms session):

Main pipeline run 36714890997 failed in e2e shard 1:

• derived-columns.spec.ts:317 ("a computed column can be filtered on…") failed on all three attempts, with three different symptoms: a column load timed out after 10 s, a waitForURL timeout, and a missing row.
• kanban.spec.ts:172 was flaky.
• Page errors show ClientDb unavailable … local database has not finished opening.

This PR changes only CLAUDE.md, CONTRIBUTING.md, browser/CONTRIBUTING.md, browser/README.md and a planning note, so it can't cause a table e2e failure. The Downstream job was green.

I merged develop in (eee8c84, bringing in dacd78d) and dispatched the pipeline once more. If that e2e fails again the same way, I'll treat it as a real flaky-test problem outside this PR and report it separately instead of re-running again.

---

Generated by Claude Code

[joepio]

Trekmeester: included in batch #1915 (run 36779344325). It lands with it when green; if this PR's own tests fail there, I take it out and say why here.

[joepio]

Trekmeester: included in batch #1947 (run 36781515104; #1915 had been merged without it). It lands when that run is green; if this PR breaks something there, I take it out and say why here.

[joepio]

Merged into develop through batch #1947 (squash 1f892aa); this head was verified as part of it. Closing.

---

Generated by Claude Code