Support more chain types, implement vector search powered by huggingface inference api

[logancyang]

• Added ChainFactory for more singleton chain instances of different types
• Added logic for vector search powered by Huggingface Inference API, will ship this in the next minor release.

Problems encountered:

• Streaming causes ConversationalRetrievalQAChain to always start answers with a rephrased question Streaming causes LLM to always start answers with a rephrased version of the question langchain-ai/langchainjs#754
  • Need to wait and test more and wait for this fix to launch "Use Active Note as Context" for long notes.
• Refused to set unsafe header "User-Agent" error, happens only when there's non-streaming ChatOpenAI. This is unfortunate because LangchainJS is using openai-node which causes this error. The recommended way is to have a server that uses LangchainJS and openai-node, but this is a pure client app. Avoid non-streaming for now.