Fix glob recursion and streamline embedding defaults

[mmprotest]

Summary

• ensure recursive indexing works with glob patterns by updating file discovery
• add a lightweight hashing embedding backend, make sentence-transformers optional, and expose backend selection in the CLI/server
• stream vector search over database vectors to avoid loading the entire table

Testing

• pytest

---

https://chatgpt.com/codex/tasks/task_e_68e02dca342c8331aca6ceab0db428d9