Fix remote vector model embedding requests

[codedogQBY]

Summary

• Route remote embedding API requests through the platform fetch layer so desktop uses Tauri HTTP instead of WebView fetch.
• Share remote embedding request parsing/dimension detection across settings, onboarding, and vectorization.
• Keep OpenAI-compatible and Ollama embedding response handling in one core helper.

Fixes #217

Test plan

• pnpm exec biome check packages/core/src/rag/embedding-request.ts packages/core/src/rag/index.ts packages/core/src/rag/vectorize-trigger.ts packages/core/src/rag/embedding-service.ts packages/app/src/components/settings/VectorModelSettings.tsx packages/app/src/components/onboarding/steps/EmbeddingPage.tsx packages/app-expo/src/components/onboarding/steps/EmbeddingPage.tsx
• pnpm --filter @readany/core exec tsc --noEmit
• pnpm --filter app exec tsc --noEmit
• pnpm --filter @readany/app-expo exec tsc --noEmit
• pnpm --filter @readany/core test