Implement M1 authoring loop: library, editor, live preview, persistence

M1 MVP from docs/IMPLEMENTATION-PLAN.md, with the /alignment pass applied.

- core: Snippet model + factory; prepareSpecForRender (copy-not-mutate); per-theme Vega chart config
- state/orchestration: SnippetStore with debounced auto-save; IndexedDB adapter + read-time migration; startup hydration + write-through persistence
- ui: SnippetLibrary, SpecEditor (Monaco edcore.main — full editor features, JSON-only languages), LivePreview
- build: Monaco/Vega manual chunks; raised PWA precache ceiling
- alignment: flush a valid draft on snippet switch (+regression tests); TODO breadcrumbs for the preview render race and the window.confirm delete
- housekeeping: gitignore .claude/projects/

docs/architecture/01-state-and-stores.md

@@ -14,6 +14,14 @@ not components" architecture, and it carries no build-time magic. The principles
 (one source of truth, derive-don't-duplicate, actions outside components, thin components)
 are the durable part — they would survive a change of library.
 
+**Lineage (why React + Zustand).** The UI began on Preact + `@preact/signals` and migrated to
+**React + Zustand** at M0, before feature work. The driver was *React-ecosystem friction* —
+real-React-only libraries not cooperating with `preact/compat` — **not** the signals model.
+Switching framework while nothing was implemented yet was also the one cheap moment to pick
+the lowest-migration-risk state library, so signals gave way to Zustand. A bonus: borrowing
+from the React-based vega/editor reference (see [08](08-vega-editor-techniques.md)) then ports
+directly rather than through `preact/compat`.
+
 ---
 
 ## 1. The Primitives