Support for OpenAI-compatible embedding endpoints

Several self-hosted embedding servers expose OpenAI-compatible `/embeddings` endpoints:
- [Infinity](https://github.com/michaelfeil/infinity)
- [vLLM](https://github.com/vllm-project/vllm)
- [Ollama](https://ollama.ai/)

The current OpenAI provider has a hardcoded `API_BASE_URL`. A configurable `base_url`
would make all these work without needing separate provider implementations.

**Possible API**
```python
client = Client()
response = client.embed(
model="BAAI/bge-small-en-v1.5",
inputs=["Hello world"],
base_url="http://localhost:7997" # or via env var
)
```

Since the request format is already OpenAI-compatible, only the URL routing would change.

Thanks for catsu, the unified API and model metadata catalog are really useful!