fix(rag): route remote embeddings through platform fetch

[codedogQBY]

Summary

• Add a shared remote embedding request helper that uses getPlatformService().fetch() by default, falling back to browser fetch only when the platform service is unavailable.
• Route book vectorization remote embedding batches through the helper so desktop uses Tauri HTTP instead of WebView fetch.
• Route the desktop Vector Model settings “Test” request through the same helper so local Ollama endpoints can be validated without CORS preflight failures.

Analysis

Issue #398 happens because remote embedding requests to local Ollama were made with WebView/browser fetch from the tauri.localhost origin. Ollama rejects that CORS preflight, so ReadAny reports failed to fetch even though the local service is reachable.

The desktop app already registers TauriPlatformService, whose fetch implementation uses @tauri-apps/plugin-http and is not subject to WebView CORS. This change moves remote embedding requests onto that platform abstraction while preserving the existing OpenAI-compatible and Ollama /api/embed request/response shapes, batch truncation, and per-chunk retry behavior.

Fixes #398

Verification

• git diff --cached --check
• pnpm exec biome check packages/core/src/rag/remote-embedding.ts packages/core/src/rag/remote-embedding.test.ts packages/core/src/rag/index.ts packages/core/src/rag/vectorize-trigger.ts packages/app/src/components/settings/VectorModelSettings.tsx
• pnpm --filter @readany/core test -- remote-embedding
• pnpm --filter @readany/core exec tsc --noEmit
• pnpm --filter app exec tsc --noEmit