refactor(login): one entry point for first run and add-account

[dmnyc]

Adding a second account presented LoginView — its own screen — while a cold launch presented SplashView with a bottom sheet. Two hand-written copies of "choose an account", and they had drifted apart in three ways that cost real functionality.

Supersedes #446. That PR's commit is included here rather than rebased away, so the Cancel fix survives LoginView being deleted instead of going with it. #446 can be closed unmerged.

The three gaps in add-account

No watch-only. NostrLoginSheet.login() accepts an npub / nprofile via Nip19.decodeNostrUri and calls NostrKey.saveWatchOnly. LoginView.login() only ran parseNsec — its placeholder read nsec1... and its error didn't mention npub — so a watch-only second account couldn't be added at all. Multi-account supports it fine: saveWatchOnly → save → addToAccountList.

No Apple or Google. Add-account offered nsec only, in the same build that shows both buttons at first launch (gated on AppleAuthConfig.isConfigured / GoogleAuthConfig.isConfigured). A second account whose key lives in iCloud Keychain or Google had no way in.

No trimming. The sheet trims pasted input; LoginView passed the raw string to parseNsec. An nsec with a trailing newline — i.e. from any password manager — failed on one screen and worked on the other.

Plus the cosmetic divergence you'd expect: different logo, title, button shape, and error copy.

The change

SplashView was already callback-shaped and holds no first-run-only state, so it takes a Mode instead:

• .firstRun — unchanged behavior, including the 1.8s curtain.
• .addAccount — shows Cancel, skips the curtain.

The curtain hides the launch animation's safe-area settling. Presented over a running app the safe area is already stable, so keeping it would mean 1.8s of blank screen after tapping Add account.

ContentView grows one loginEntry(mode:) carrying the Nostr sheet and the Google / Apple covers, plus a shared finishLogin. What actually differs between the two entry points turns out to be two lines — dismissing the cover, and the shorter loading delay an account switch gets.

Net −288 / +122, and the three gaps close for free: add-account is now literally the same code path as first launch.

Decisions worth a look

Apple / Google are offered in add-account. They're how a cloud-backed key gets in, so excluding them would leave the gap half-closed. It does mean someone can add an account that's already signed in elsewhere. Easy to gate to .firstRun if you'd rather.

The avatar collage stays in both modes. It costs a profile fetch and a metrics socket per open, but it is the screen's design — without it, add-account is a bare screen with three buttons, which isn't "one view" in any useful sense. viewModel.cancel() on disappear already tears both sockets down.

LoginView.swift is deleted along with its four project.pbxproj references — it lived at the repo root, which is individually listed rather than a fileSystemSynchronizedGroup. The build passing is the check that the right four lines went.

Testing

Built and installed on the simulator. Two things worth a reviewer's attention, since they're what a refactor like this would break: that first launch still settles behind the curtain, and that Apple's brand-new-account path still routes into the sign-up wizard rather than the loading screen — that branch in finishLogin is the only routing I restructured rather than moved.

[barrydeen]

Reviewed; the 'two paths differ by two lines' claim checks out line by line.

• First-run: every old splash branch (Nostr onLogin, Google onDone, Apple new/restored) maps 1:1 onto loginEntry(mode: .firstRun) + finishLogin, including the only restructured branch — Apple brand-new still sets keypair/signUpExistingKeypair and routes to the sign-up wizard. accountSwitchInProgress remains add-account-only; onboarding-first ordering preserved both directions.
• Add-account: old LoginView callback semantics reproduced exactly (dismiss cover, switch-delay flag only when already onboarded).
• LoginView.swift and all four pbxproj refs gone with zero surviving code references; the sole other SplashView() caller is the #Preview, compiling via the defaulted mode.
• The three gaps (watch-only npub, Apple/Google, trim) close for free since NostrLoginSheet becomes the single nostr path; #446's Cancel fix survives.
• Mode-gated curtain skip is right — the 1.8s hold exists only for launch-animation safe-area settling.

Non-blocking nits:

1. Cancel hardcodes .foregroundStyle(.black) on the white capsule — fine on the splash design, but one of the few hardcoded colors here.
2. mode: Mode = .firstRun defaults silently to the uncancellable launch variant for any future call site; a required parameter is the safer contract.
3. Apple/Google in add-account (allows adding an account signed in elsewhere) is the product call the PR flags — fine to keep.

Note: no CI on the branch and no iOS toolchain at the reviewer's machine; build correctness relies on the author's simulator verification (and the pbxproj deletion check is 'build passes', which I couldn't run).