Refactor: split app.svelte.js (2024 lines) into focused stores

[silverbucket]

`src/lib/stores/app.svelte.js` is 2024 lines covering connection lifecycle, account swap, generation orchestration, song editor, band editor, import/export, migrations, and config. Each section is independent state with its own concerns.

The Svelte 5 `$state` pattern doesn't require a god-store. Multiple stores composed via `setContext("app", {...connection, ...songs, ...band})` keeps cohesion high.

Fix (incremental):

1. Extract migrations/import/export → `stores/data-io.svelte.js`
2. Extract account/connection lifecycle → `stores/connection.svelte.js`
3. Extract band-editing → `stores/band.svelte.js`
4. Extract song editor → `stores/song-editor.svelte.js`
5. Leave generation+roll in the main store.

Do this gradually — one extraction per PR, verify tests pass.

(From comprehensive review, finding E1.)

[silverbucket]

Triage

Still applicable. `src/lib/stores/app.svelte.js` is now 2429 lines (was 2024 at issue time — slight drift from #41-50 fixes and unrelated work).

Importance: 3 (low). No bug, no user-visible impact. Pure maintainability investment.

The proposed extraction order in the issue (data-io → connection → band → song-editor) is reasonable. Worth pairing each extraction with #72 (underscore-prefix convention) and #73 (toast helpers) since both tighten the API surface in ways that make extraction cleaner.

I'd add one note: the `$state` and `$effect` runes can move freely across modules in Svelte 5, but `setContext("app", ...)` is the integration point — stick with a single context object rather than per-store contexts to keep components agnostic of where state lives.