fix: cancel abandoned highlight() semantic work and harden WASM engine recovery

highlight() used Promise.race with a 10s timeout, but the losing
semantic phase promise continued running 25 WASM micro-batches,
saturating the event loop and degrading all subsequent operations
(find() going from ~200ms to ~10,000ms).

Add AbortController to highlight() so the semantic phase stops
immediately on timeout or error. Pass abort signal through
embedBatch() → EmbeddingManager → micro-batch loop.

Also add defensive hardening:
- CandleEmbeddingEngine: try/catch around WASM calls resets engine
  state on failure so next call triggers re-initialization
- WASMEmbeddingEngine: initialize() now checks underlying Candle
  engine state, not just its own flag, completing the recovery chain

src/embeddings/EmbeddingManager.ts

@@ -219,7 +219,7 @@ export class EmbeddingManager {
    * @param texts Array of strings to embed
    * @returns Array of embedding vectors (384 dimensions each)
    */
-  async embedBatch(texts: string[]): Promise<number[][]> {
+  async embedBatch(texts: string[], options?: { signal?: AbortSignal }): Promise<number[][]> {
     if (texts.length === 0) return []
 
     const isTestMode =
@@ -248,6 +248,10 @@ export class EmbeddingManager {
     // so other requests (HTTP, timers, I/O) can proceed
     const allResults: number[][] = []
     for (let i = 0; i < texts.length; i += MICRO_BATCH_SIZE) {
+      if (options?.signal?.aborted) {
+        return allResults
+      }
+
       const batch = texts.slice(i, i + MICRO_BATCH_SIZE)
       const batchResults = await this.engine.embedBatch(batch)
       allResults.push(...batchResults)