Adding semantic HTML chunking using Unstructured.io

[tarockey]

Still in progress, but adding semantic HTML chunking.

The strategy should apply to the rest of the document chunkers.

Overall method:

1. use Unstructured to chunk by title
2. use an embedding model to embed each split chunk NOTE: currently using spacy by default. open task to add configurable embedding model.
3. compare split chunks to find and combine semantically similar chunks

Changes:

• Added SEMANTIC as a chunking strategy to configuration
• Added SEMANTIC_SIMILARITY_THRESHOLD to configuration (this defines how semantically similar two chunks must be, before they are combined)
• Updated load_documents to accept a config as a parameter, to allow these params to be passed, without overloading load_documents
• updated html_loader to include semantic chunking logic.

[julia-meshcheryakova]

• [x] Do not merge until prerelease in merged into development. #556