fix(desktop): carry typed OAuth failure reasons from main-process producers

[orangeCatDeveloper]

Summary

subscriptionResultMessage localized OAuth login failures by matching English Host prose with regexes. One of the three regex families — did not present OAuth|no matching OAuth presentation — described failures Desktop produces itself: the presentation timeout and the absent-request throw in runtime-host-oauth-presentation.ts. Matching your own prose is a contract you cannot see break; rewording either sentence silently drops the branch.

Those producers now throw a typed OAuthPresentationError, and get-auth-url maps it alongside the Host's operation codes onto the failure envelope:

• operation_conflict → login_in_progress
• OAuthPresentationError → presentation_failed
• operation_unavailable → experimental_disabled (unchanged)

The renderer catalog maps the two new reasons per locale and the presentation regex is deleted. The enrollment is disabled and already in progress prose fallbacks stay: the first covers Hosts older than these codes, and the second still has a live producer in presentation.expect, which throws a plain error. The ad-hoc loginConflict / browserPresentFailed keys fold into resultCodes with the same wording.

Refs #2672

Verification

apps/desktop full dist suite        2489 pass / 0 fail
apps/desktop typecheck (4)          0 errors
renderer architecture ratchet       passes against the base
npm run format:check                clean

Not run: Electron e2e against a live Host login flow.

AI use

• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code — traced the producer→presenter path, implemented the typed error and reason mapping, removed the regex, added the tests, and wrote this description.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — login-conflict and browser-presentation failures render catalog copy through typed reasons instead of English prose matching.

[Astro-Han]

Thanks for replacing presentation-message matching with structured reasons. The architecture looks appropriate: Desktop owns browser presentation failures, and Runtime Host continues to own the OAuth attempt and its terminal state. I found one overly broad Host-code mapping, noted inline.

There is also a small contract simplification worth making across this change: add the two new reasons to the existing SubscriptionActionFailureReason used by the preload and renderer ports, then derive the main-process and catalog types from that contract. Currently the new main-process union and renderer string extensions describe values missing from the declared IPC result contract. Sharing that existing contract would remove duplication and let the compiler check the producer-to-consumer path.

Reviewed at c27e9962c1ee81ea1caf26366a03ebbe08aa287b; current CI is passing. The full suite and a live Electron/Host OAuth flow were not rerun. A focused test carrying a real browser-open rejection through waitForPresentation to the IPC result would cover more than injecting a presentation error from startOAuthLogin.

AI assistance: Codex with a deep reviewer; the finding was cross-checked against the Host producer and renderer consumer.

[Astro-Han]

Thanks for the update. An independent fresh-eye re-review of e3ffc3116bf5bc03264683a9fb2fbcd198f67b21 found no remaining P0–P3 issue.

The previous conflict-mapping finding is fixed: only the local OAuthLoginInProgressError becomes login_in_progress; a generic Host operation_conflict, including catalog capacity exhaustion, remains unknown. The failure reasons now also share the existing Core contract rather than separate main/renderer unions. The new browser-open failure regression follows the presentation promise into the IPC handler and checks Host cancellation.

The ownership boundary remains sound, and I do not have a further simplification to request. One small description update: the PR summary still says operation_conflict → login_in_progress and that presentation.expect throws a plain error; those statements describe the older implementation and should match the corrected code.

Current CI is passing. This review checked the exact source and regression paths; a separate local Desktop build could not complete because dependent workspace build artifacts were absent, so it is not claimed as a passing local test run. A live Google/Host OAuth flow was not repeated.

AI assistance: a new Codex deep reviewer performed the fresh-eye pass, followed by a primary-agent cross-check.

[Astro-Han]

Thanks for the fixes. Approved at e3ffc3116bf5bc03264683a9fb2fbcd198f67b21: the independent fresh-eye review found no remaining code findings, the previous conflict-mapping defect and duplicated failure contract are resolved, and current CI is passing.

AI assistance: Codex deep review with a primary-agent cross-check.