feat: typed login() rejection shape across platforms (+ useAssetLinks plist passthrough docs/constant)

[airowe]

Implements #110 — a documented, typed error shape for login() rejections across platforms — plus companion plumbing for frontegg/frontegg-ios-swift#293.

What

Typed login errors (commit 1):

• TS exports: FronteggLoginErrorCode ('user_cancelled' | 'oauth_failed' | 'network' | 'unknown'), FronteggLoginError { code, message, userCancelled, nativeCode?, nativeMessage? }, normalizeLoginError(); login() rethrows the normalized shape so consumers get it even against older native layers. Documented in docs/usage.md.
• iOS: fixes a latent bug — the current rejecter(error.failureReason, …) passes nil (failureReason only exists on the inner FronteggError.Authentication enum), so JS always received EUNSPECIFIED. New normalizedLoginError(_:) switches on the real error cases (verified against FronteggSwift 1.3.13, the podspec pin): .operationCanceled → user_cancelled; the OAuth family → oauth_failed; .networkError → network; conservative unknown otherwise, raw reason preserved in userInfo.
• Android: the module already forwards the SDK's login callback into the promise (FR-25938) but rejects everything as flat LOGIN_ERROR; now normalized — CanceledByUserException → user_cancelled, FailedToAuthenticateException → oauth_failed, IOException → network — with the raw exception class/message preserved.

useAssetLinks passthrough (commit 2): the wrapper passes no config to native init (FronteggSwift decodes Frontegg.plist itself), so a JS option would be dead plumbing. Instead: docs/setup.md documents the plist key as the passthrough (Codable ignores unknown keys today; takes effect automatically if/when #293's native option ships), and getConstants() now exposes useAssetLinks (typed via a new FronteggConstants interface), matching how Android surfaces its config.

Validation

• yarn test: 8/8 (5 new tests covering the normalization paths)
• yarn typecheck / eslint: clean vs. baseline (the two example/ errors and one warning exist identically on master)
• Android: :frontegg_react-native:testDebugUnitTest BUILD SUCCESSFUL via example/android (module Kotlin + new JVM tests)
• iOS: swiftc -parse clean; the error-mapping function additionally typechecked against the real FronteggSwift 1.3.13 error model. A full pod-install host-app build is the remaining verification — draft until we've done that.

One mapping judgment call flagged for review: iOS .authError(.other(Error)) may wrap URLError network failures; we left it unknown (raw reason preserved) rather than guess.

[dianaKhortiuk-frontegg]

Approving. Verified the JS layer locally, because CI did not run it.

Only Android E2E and iOS E2E reported on this PR — Install | Build | Test and Lint and Typecheck never ran, so the 77 lines of new tests and the new TypeScript types were unverified. Ran them against the PR head:

• jest — 8/8 pass, including login resolve/reject and all normalizeLoginError cases (stable codes, unknown-code passthrough, null/non-error rejections).
• tsc --noEmit — no errors in src/.
• eslint — 0 errors. The one no-shadow warning on FronteggNative.ts:177 is pre-existing on master and untouched here.

The fix itself is right: the iOS bridge rejected with error.failureReason, which is nil on the outer FronteggError (only the inner Authentication enum implements it), so every iOS login failure surfaced as EUNSPECIFIED. Android's mapping is unit-tested for all four codes.

Two follow-ups, neither blocking:

1. The iOS mapping has no unit test — this repo has no Swift unit-test target (only ReactNativeExampleUITests), so there's nowhere to put one. Worth adding that target separately.
2. The missing CI jobs are worth investigating — a PR that changes JS/TS should not merge without the JS test and typecheck jobs running.