feat(TRUEREF-0007): implement pluggable embedding generation and vector storage

Add EmbeddingProvider interface with OpenAI-compatible, local (optional
@xenova/transformers via dynamic import), and Noop (FTS5-only fallback)
implementations. EmbeddingService batches requests and persists Float32Array
blobs to snippet_embeddings. GET/PUT /api/v1/settings/embedding endpoints
read and write embedding config from the settings table.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/lib/server/embeddings/embedding.service.ts

@@ -0,0 +1,101 @@
+/**
+ * EmbeddingService — batches embedding requests and persists results to
+ * the snippet_embeddings table.
+ */
+
+import type Database from 'better-sqlite3';
+import type { EmbeddingProvider } from './provider.js';
+
+interface SnippetRow {
+	id: string;
+	title: string | null;
+	breadcrumb: string | null;
+	content: string;
+}
+
+const BATCH_SIZE = 50;
+const TEXT_MAX_CHARS = 2048;
+
+export class EmbeddingService {
+	constructor(
+		private readonly db: Database.Database,
+		private readonly provider: EmbeddingProvider
+	) {}
+
+	/**
+	 * Embed the given snippet IDs and store the results in snippet_embeddings.
+	 *
+	 * Only snippets that actually exist in the database are processed.
+	 * Results are upserted (INSERT OR REPLACE) so re-embedding is idempotent.
+	 *
+	 * @param snippetIds - Array of snippet UUIDs to embed.
+	 * @param onProgress - Optional callback invoked after each batch with
+	 *                     (completedCount, totalCount).
+	 */
+	async embedSnippets(
+		snippetIds: string[],
+		onProgress?: (done: number, total: number) => void
+	): Promise<void> {
+		if (snippetIds.length === 0) return;
+
+		const placeholders = snippetIds.map(() => '?').join(',');
+		const snippets = this.db
+			.prepare<string[], SnippetRow>(
+				`SELECT id, title, breadcrumb, content FROM snippets WHERE id IN (${placeholders})`
+			)
+			.all(...snippetIds);
+
+		if (snippets.length === 0) return;
+
+		const texts = snippets.map((s) =>
+			[s.title, s.breadcrumb, s.content]
+				.filter(Boolean)
+				.join('\n')
+				.slice(0, TEXT_MAX_CHARS)
+		);
+
+		const insert = this.db.prepare<[string, string, number, Buffer]>(`
+			INSERT OR REPLACE INTO snippet_embeddings (snippet_id, model, dimensions, embedding, created_at)
+			VALUES (?, ?, ?, ?, unixepoch())
+		`);
+
+		for (let i = 0; i < snippets.length; i += BATCH_SIZE) {
+			const batchSnippets = snippets.slice(i, i + BATCH_SIZE);
+			const batchTexts = texts.slice(i, i + BATCH_SIZE);
+
+			const embeddings = await this.provider.embed(batchTexts);
+
+			const insertMany = this.db.transaction(() => {
+				for (let j = 0; j < batchSnippets.length; j++) {
+					const snippet = batchSnippets[j];
+					const embedding = embeddings[j];
+					insert.run(
+						snippet.id,
+						embedding.model,
+						embedding.dimensions,
+						Buffer.from(embedding.values.buffer)
+					);
+				}
+			});
+			insertMany();
+
+			onProgress?.(Math.min(i + BATCH_SIZE, snippets.length), snippets.length);
+		}
+	}
+
+	/**
+	 * Retrieve a stored embedding for a snippet as a Float32Array.
+	 * Returns null when no embedding has been stored for the given snippet.
+	 */
+	getEmbedding(snippetId: string): Float32Array | null {
+		const row = this.db
+			.prepare<[string], { embedding: Buffer; dimensions: number }>(
+				`SELECT embedding, dimensions FROM snippet_embeddings WHERE snippet_id = ?`
+			)
+			.get(snippetId);
+
+		if (!row) return null;
+
+		return new Float32Array(row.embedding.buffer, row.embedding.byteOffset, row.dimensions);
+	}
+}