Parent child retriever examples

[joemcelroy]

This PR demonstrates:

• Example 1: Using nested dense vector for document -> Full document + nested multiple passages.
• Example 2: Using nested dense vector for document -> chunked document + nested smaller passages.

Fixed couple issues on:

• remove unneeded function in bm25 example
• remove shard settings so can run on serverless

[miguelgrinberg]

Comments:

• Not including the commands to upload the model to the ML node means that most people will not be able to run this notebook to completion.
• In 8.11 you do not need to specify the vector dimensions or the index variable. Actually I think with 8.11 you can get away w/o declaring any properties on the nested field.
• At some point you had reservations with using dot product for the minilm-l6-v2 model. I think it is fine based on the information here, but I mention it just in case you prefer to switch to cosine.
• I found the big and small chunking difficult to understand. In reality, when you create the larger chunks you do not preserve a link to the complete document, so these large chunks are effectively parent documents in their own right. So I'm not sure doing a two-level splitting contributes much to the example. Both examples could use complete doc and nested passages, with example 1 showing how to get only the passages, and example 2 showing how to get the parent and the passages. No need to change anything, just thinking out loud here that I think I would have found it easier to follow without the large chunks...
• We have a mix of helpers.bulk and client.bulk in our notebooks. We should clean this up at some point.

[joemcelroy]

Hey @miguelgrinberg

Going to do the following updates:

• add the part on how to load the ML model into the notebook so can be run to completion
• update not declaring the nested field
• remove the 2nd parent / child chunking example. Not sure if its helping here. Keep the notebook simple IMO.