Langroid

* Remove all dependencies on Haystack
* Make transformers an extra (eliminates torch dependency); use `pip install "langroid[transformers]"` to get it
* Make `unstructured` library optional + extra; use `pip install "langroid[unstructured]"`) to get it
* Update ChromaDB to latest 0.4.23
* Default DocChatAgent to OpenAI Embeddings (eliminates dependency on transformers in core-langroid).
- To use hugging-face embeddings, install langroid with `pip install "langroid[hf-embeddings]"`

* Task: fix SEND_TO handling
* json tool detection: handle weak LLM producing unquoted string values
* Misc minor fixes, example tweaks

SQLChatAgent: handle multiple schemas

Make batch utilities available via lr:

- `run_batch_tasks`
- `llm_response_batch`
- `agent_response_batch`

E.g. you can now do:


import langroid as lr
...
lr.run_batch_tasks(...)
...

Minor:
* fix in Task.py: When interactive = True, set agent.default_human_response = None
* openai_gpt.py - when getting available models, catch exception in case there's an OpenAI outage, so we don't fail when we only want to use a local LLM

Support `ollama` [OpenAI API-compatility](https://ollama.com/blog/openai-compatibility), i.e ollama LLM server now mimics the OpenAI API, so any code that used to work for OpenAI LLMs will now work with a simple change of `api_base`.

Langroid takes care of setting the `api_base` behind the scenes, when you specify the local LLM using `chat_model = "ollama/mistral"`, e.g.

python
import langroid.language_models as lm
import langroid as lr

llm_config = lm.OpenAIGPTConfig(
chat_model="ollama/mistral:7b-instruct-v0.2-q8_0",
chat_context_length=16_000, adjust based on model
)
agent = lr.ChatAgent(llm=llm_config)
...


See more in this tutorial on [Local LLM Setup with Langroid](https://langroid.github.io/langroid/tutorials/local-llm-setup/)