refactor(core): explicit per-field operation variants + deprecate renderMarkdown

[LeslieOA]

Summary

Two API hygiene changes bundled into one PR — both small, both touching the public surface.

1. Replace Partial<CanvasNode> with explicit operation variants (closes #5)

The last FFI-hostile shape from the original audit. Partial<T> has no clean Rust equivalent — it'd force a stringly-typed HashMap<String, JsonValue> or a sparse-field struct across JSI / wasm-bindgen. Replaced with six per-field operation variants:

• updateNodeColor(nodeId, color)
• updateTextNodeText(nodeId, text)
• updateLinkNodeUrl(nodeId, url)
• updateFileNode(nodeId, file, subpath)
• updateGroupNodeLabel(nodeId, label)
• updateGroupNodeBackground(nodeId, background, backgroundStyle)

Each Operation variant carries the new value(s) and prev value(s) for invertOperation. Apply paths verify node type before mutating (silent no-op on missing or mismatched type, consistent with removeNode's if (!node) return pattern).

Exported GroupBackgroundStyle = 'cover' | 'ratio' | 'repeat' so consumers can type-narrow without reaching into GroupNode.

Breaking: updateNodeContent is gone from the public API. No external consumers in this repo. Workspace's canvas-ui has the same shape and will need the same migration when it next pulls from upstream.

2. Deprecate renderMarkdown (closes #26)

The prop on CanvasView was vestigial — nothing in src/renderer/** consumed it because TextNodeContent returns null and all text-node bodies render through paragraphBuilder / Skia. Marked @deprecated in both declaration sites (CanvasView.tsx, CanvasContext.tsx) with a JSDoc explaining what actually drives text rendering. README's "What's inside" section gains a Markdown rendering subsection enumerating supported features (inline marks, blocks, GFM extensions) and what isn't (anything needing a React overlay).

No behavioural change — the prop continues to be accepted and ignored. Removal is a separate decision for a later major version.

Verification

• npm test: 38/38 (tests reference state.history.canRedo which is unchanged)
• npm run typecheck: clean
• npm run lint: exit 0

Why bundle these

Both are public-API surface hygiene, both small, both serve the same forward direction (cleaner shape for a future Rust core, clearer contract for consumers). Splitting into two PRs is overhead for ~285 LOC across 6 files. If you'd rather they were split, easy to do.

What's next after this merges

• Closes refactor(core): replace Partial<CanvasNode> in updateNodeContent with explicit operation variants #5 (last FFI-hostile API shape — clears the runway for feat(core): port pure-logic core to Rust (JSI native + wasm-bindgen web) #6 Rust port)
• Closes chore(renderer): deprecate renderMarkdown prop and document actual markdown support #26 (renderMarkdown clarity)
• Frees the project board to pick the next thread — most likely chore: set up CI (typecheck + lint + test on PR) #8 (CI) or one of the Canvas Candy issues

Test plan

• Library checks (test / typecheck / lint)
• Smoke test in playground (no behavioural change expected — updateNodeContent wasn't called anyway)

Refs: #5, #26