Langroid

feat: Add Gemini Embeddings, see [docs](https://langroid.github.io/langroid/notes/gemini/)

Thanks abab-dev!

fix: `openai_gpt.py` - don't extract reasoning when msg is None.

feat(minor): extract reasoning from LLM response if within thought-delimiters

Some LLM APIs include the reasoning part within delimiters like `<think> ... </think>`, rather than in a separate `reasoning` field.
The delimiters are configurable via `LLMConfig.reasoning_delimiters: Tuple[str,str]`.
This feature extracts such reasoning and puts it in the `reasoning` field of the Langroid `LLMResponse` object.

See example `examples/reasoning/agent-reasoning.py` and try with `ollama/deepseek-r1:32b` (the full R1 API does
give a separate `reasoning_content` field in the API result, but the distills do not).

fix: minor fixes in `SQLChatAgent` and `openai_gpt.py`

- `SQLChatAgent` - remove redundant `available_tool_names()` method
- `openai_gpt.py` - handle case where `"choices"` is None - happens with `gemini-2.0-flash-thinking-exp`

chore: `VectorStore`: move `embedding_fn`, `embedding_dim` to base class.

fix: Weaviate fix bug in handling None distances