2.4.6

What's Changed

• Added optional Embedding server URL and Embedding API key settings so RAG indexing can use a separate OpenAI-compatible embedding server.
• Kept the current single-server setup as the default: blank embedding server inherits the chat/default server, and blank embedding key inherits the chat/default key.
• Added support for not-needed on the embedding API key to suppress the Authorization header.
• Updated chat and embedding model browsing so each picker queries the correct effective server.
• Stored embedding metadata now records the effective embedding server and model without storing API keys; old indexes remain loadable and still recommend re-indexing when the embedding backend changes.

Verification

• npm run build
• git diff --check
• GitHub Actions release workflow passed and uploaded main.js, manifest.json, and styles.css.