This PR has no intention to merge. but interesting features sharing.

I don’t really expect this PR to be merged, but I wanted to share a few ideas and implementations that I found interesting and potentially useful. While using the project personally, I noticed several areas that could be improved or extended, so I put this together (with significant help from Copilot / Opus 4.7). 1. Treat multiple TTL files in a directory as a registry, with support for loading and unloading them dynamically. 2. Compile each TTL to improve load times, and automatically offload unused ones to disk after a certain period. This is particularly useful when dealing with very large ontologies, where load times can be significant and memory usage can become excessive. 3. Add DuckDB as an additional backend option alongside PostgreSQL for data source integration. 4. Introduce configurable default options to make setup and tuning easier. 5. Add an OpenAI-compatible embedding layer to allow integration with external embedding APIs. Even if this doesn’t get merged, I hope it can serve as a useful reference or spark discussion about possible directions. Also, I want to say that I’ve been using open-ontologies extensively and really appreciate the work—thank you!