fix(platform): name a failed import status read on the OAuth apps card

[yannickmonney]

What changed

The Settings > Connectors OAuth apps card asks the Knowledge import lanes (cloud_import/queries:getOauthAppStatus for OneDrive and for Google Drive's import half) whether a deployment app stands behind their rows. On main, a read that settled in error (data: undefined, isLoading: false, e.g. a 503) collapsed envConfigured to false. The row then said Not configured and offered Configure (and Use Entra ID SSO app on OneDrive), with no failure named and no retry. An admin could override a deployment app that was there.

oauth-apps-card.tsx now reads both status queries through readStateOf, the same pattern #4378 used for the same endpoint in the Documents connect dialog:

• Row status. A row whose status that read decides, and which neither the org's app list nor the connector catalog already answers, gets statusUnavailable. It shows a Status unavailable badge and no Configure or Use Entra ID SSO app.
• Alert. One CatalogLoadError names the failure (connectors.oauthApps.statusLoadFailed). Its Try again refetches only the failed status reads, and failureKey makes each new failure announce again.
• During a retry. The row stays put: loading leaves out a read that is unavailable, so the card doesn't re-mask. Try again is busy, keeps its node and its focus, and keeps them again if the retry fails too.
• After recovery. A retry that works removes the alert, and a focused Try again hands its focus to the card (SettingsSection with tabIndex={-1}). Configure comes back only once a read answers: configured: false → Not configured, source: 'env' → Deployment app.
• Rows known from elsewhere keep their answer: an org row from the app list, or Google Drive when the catalog says its connector lane has an env app.
• App-list alert: unchanged.

Also in this PR:

• Copy: EN/DE(du)/FR(tu) for statusUnavailable and statusLoadFailed. Neither key contains ß, so de-CH needs no override.
• Docs: one sentence in docs/{en,de,fr}/platform/admin/connectors.md.
• Register: a paragraph in tests/manual/reference/automation.md, placed right after the bug(platform): Cloud import connect treats a failed app-status read as not connected #3864 connect-dialog paragraph.

How I verified it

• New suite. oauth-apps-card.status-read.test.tsx runs the real hooks, adapters, backendFetch and retry policy against syntheticBackend(). It passes 13/13 in jsdom, covering:
  • a failed OneDrive read and a failed Drive import read;
  • rows known from the app list or the catalog;
  • a held retry: the row holds, Try again is busy and keeps focus, only the failed read refetches, then focus moves to the card;
  • a retry that fails again;
  • both reads failing behind one alert;
  • configured: false, an env source, and the loading mask;
  • EN/DE/FR;
  • an axe pass with the unavailable row and the alert on screen (second commit).
• Red on main. With main's oauth-apps-card.tsx, 9 of 13 fail. The 4 that pass are the answered or known-row cases.
• Real Chromium 1194. The 12 tests of the first commit and the existing oauth-apps-card.test.tsx pass 22/22 in headless Chromium through a review-only Vitest config outside the clone, so the focus handoff was checked in a real browser too. The axe test was added after that run.
• Related suites:
  • the connectors settings components (4 files): 61/61
  • lib/i18n (--project server): 188/188
  • docs docs.test.ts and locale-tree.test.ts: 25/25
  • bun run lint:manual: ok
  • the error-message and single-failure-toast guards: 75/75
• Static checks:
  • scoped type-aware oxlint over the card and both card suites: 0 warnings, 0 errors
  • oxfmt --check: clean
  • scoped tsc over the card, its suites, connectors-settings.tsx and the ambient files: exit 0, 0 errors
• Merge check. git merge-tree against all 53 open PR heads that touch automation.md, messages/{en,de,fr}.yml, the card or the connectors docs: each PR conflicts in the same files with this branch as with main, so the branch adds no new conflict.
  • fix(platform): report automation history read failures #4258, fix(platform): reconcile model access after failed saves #4266, fix(platform): surface legal-hold list read failures #4276 and fix(platform): preserve contact palette search at its destination #4319 already conflict with main over the whole prose block (main lines 9 to about 172), so the paragraph falls inside those existing hunks.
  • Moving the paragraph to the end of the block would add a new conflict with fix(platform): surface recent feedback read failures #4372, so I kept it next to bug(platform): Cloud import connect treats a failed app-status read as not connected #3864.
• Not run: browser E2E and the backend stack (light phase). The real disconnected-backend interaction and the browser layout remain manual.

Open points for the reviewer

• Stale refresh. A refresh that fails after an earlier answer stays silent and keeps the last answer, the same as fix(platform): name a failed cloud-import setup check, hold Connect #4378 and fix(platform): name a failed folder-list read on the Documents hub #4375.
• App-list failure. When the list read fails, rows can still claim Not configured under the existing list alert. That is out of scope here (the app-list handling is unchanged).

Closes #3893

[yannickmonney]

Independent exact-head verdict — Codex #11, separate from author Claude #6.

PASS for the scoped review; no blocking findings at 5b152b2c382653a7c7e14a4a7dc26db913e0e1b8 (+530/−22, nine files). This verdict covers that head only.

• A first getOauthAppStatus read that fails leaves the affected unknown OneDrive/Google Drive row at Status unavailable, hides Configure and OneDrive SSO reuse, and exposes Try again. Retry refetches only the failed reads behind unavailable rows. Both failures share one status alert.
• The unavailable state survives a retry, including a second failure; the retry button retains focus and becomes busy/inert. Successful recovery removes the alert and hands focus to the card. A successful configured:false, source:null restores Not configured and Configure; deployment and organization app answers remain visible with their existing configuration actions. Independently known app-list/catalog rows retain their answer.
• Initial loading masks the statuses and actions without exposing Not configured. The existing app-list error alert and its error mapping remain intact in the diff.
• EN/DE/FR docs describe the implemented failure/retry UI and match the new badge/control labels. Added DE copy uses du, FR uses tu; added DE copy has no sharp s. Both new keys exist in all three catalogs.
• git merge-tree --write-tree --name-only against all 53 other open heads touching the register, compared with fetched main 8e4c6a40b0, found no additional conflicting files or register conflict hunks. Twenty heads already conflict in the register against main; these retain the same hunk counts. This is not a claim that every older PR merges cleanly.

Observed local verification, existing dependencies only:

• Card suites: 23/23 passed (13 status-read regression tests + 10 existing tests), /opt/node/bin/node, one worker, jsdom; includes axe and retry/focus assertions. Review-only server.fs.allow handles the anticipated font restriction; workspace package aliases resolve to the reviewed checkout.
• Platform i18n: 188/188 passed, one worker.
• Docs docs.test.ts and locale-tree.test.ts: 25/25 passed, one worker.
• Scoped type-aware oxlint: three files, zero diagnostics, exit 0.
• oxfmt: both changed TSX files clean; Markdown/YAML are excluded by repository configuration. git diff --check: clean.
• Scoped TypeScript program: card, both suites, importing connectors settings component, ambient declarations and UI test setup, with the imported closure: exit 0. No whole-platform type check or suite was run.

CI: pending, stalled incident 3q1yb5m7ltvb; read-only check inspection confirmed pending jobs. No rerun, push, merge or status change. Real-browser layout and disconnected live-backend interaction were not exercised in this review. Existing stale-refresh and app-list-failure behavior remains outside this change.

Delivery fallback for TALE-316 and TALE-359 (3729d02a-eb44-42b2-a68e-1bec2cebbcd6): workspace_status returned unavailable / not_granted, so task access/comment schemas were unavailable. This single PR comment is the requested fallback for both task destinations.