Add OPFS model caching to Flare engine for instant repeat loads

[sauravpanda]

Summary

Integrate Flare's OPFS caching APIs into the FlareEngine adapter so models are cached after first download and load instantly on repeat visits.

How it works

Flare already has OPFS caching built into its WASM module:

• is_model_cached(name) — check if model exists in OPFS
• cache_model(name, bytes) — save model to OPFS
• load_cached_model(name) — load from OPFS (3-4x faster than IndexedDB)

Integration in FlareEngine.loadModel()

async loadModel(modelId: string, options?: LoadOptions): Promise<void> {
    const modelName = modelId + '.gguf';
    
    // Check OPFS cache first
    const cached = await load_cached_model(modelName);
    if (cached) {
        this.engine = FlareEngine.load(cached);
        options?.onProgress?.(1, 1); // instant
        return;
    }
    
    // Download with progress
    const bytes = await fetchWithProgress(url, options?.onProgress);
    this.engine = FlareEngine.load(bytes);
    
    // Cache for next time (fire-and-forget)
    cache_model(modelName, bytes).catch(() => {});
}

User experience

• First load: normal download with progress bar
• Second load: instant (< 100ms from OPFS)
• Show "Cached" badge on models that are already stored

Depends on

• Implement FlareEngine adapter for BrowserAI engine interface #295 (FlareEngine adapter)

Related

• Parent: Add Flare as a third inference engine backend #293

[sauravpanda]

Completed in PR #301 — OPFS caching is wired up via FlareEngineWrapper.isCached(), clearCache(), and the top-level isModelCached/deleteCachedModel/listCachedModels exports. Models cache after first download and load instantly on repeat visits.