fix: update Ollama embedding API usage

[earthenvessel]

Summary

Update to use Ollama's current embedding API.

Changes

• switch embedding requests from /api/embeddings to /api/embed
• change the request payload from prompt to input
• parse the modern embeddings response shape
• keep compatibility with legacy single-embedding responses
• return early from hybrid_search() when query embedding fails

Why

Recent Ollama versions use /api/embed for embeddings. The previous implementation was calling the older endpoint and caused document processing failures like: 500 Server Error: Internal Server Error for url: http://host.docker.internal:11434/api/embeddings

Validation

• ran python -m py_compile app.py
• verified the updated request URL and payload shape
• verified parsing for modern response format