feat: OpenSearch RAG demo

[labrenbe]

Adds a new demo that shows Retrieval Augmented Generation with OpenSearch

Stack:

• OpenSearch 3.1.0 with k-NN plugin
• Ollama running two models: nomic-embed-text:v1.5 (embeddings) & llama3.1:8b (response generation)
• JupyterLab with interactive RAG pipeline notebook

Data

• Documentation for the Stackable Data Platform crawled from docs.stackable.tech
• Pre-generated json file containing ~4200 documentation chunks (93MB, stored in git)
• Loaded into OpenSearch with a Kubernetes Job

The data file should be regenerated after each Stackable release to ensure no outdated documentation is used in the demo. This process can take more than 1 hour and is also the reason why the data is commited to git and not generated during demo installation.

Note for reviewers: You need to change the branch parameter to feat/opensearch-rag-demo in the stacks and demos files to deploy the demo. Depending on the hardware the demo should run on your local machine, but a kind cluster on Replicated with a single r1.xlarge node works best.

Release Notes

Added a new demo that shows Retrieval Augmented Generation (RAG) with OpenSearch as vector store.

[razvan]

Very nice. I didn't run the script to generate embeddings to the end but it ran fine. The demo also works.

Technically we could have a custom image with ollama, the models and our embeddings already prepared. This would reduce the amount of additional artifacts that need to be downloaded when running the demo and also simplify things.

[lfrancke]

Should this have release notes?

[labrenbe]

Should this have release notes?

I've added a release notes snippet to the PR description.