Usability improvements for samples work

[jonsequitur]

This issue collects small features and usability improvements intended to make planned sample notebooks more concise and easier to use.

Intelligent app development issues

• [x] #3194
• [ ] #3211
• [ ] https://github.com/colombod/dotnet-interactive-extension-lab/issues/38
• [x] API for token count for specific models (related PRs : #3223, #3217)
• [x] API to truncate text using token count (with optional overlapping) (related PRs : #3217)
• [x] API to split long text using token count (with optional overlapping) (related PRs : #3217)
• [x] Linq extensions for slicing and chunking (related PRs : #3217)
• [x] Search utilities over collections using similarity scores (like cosine similarity) (related PRs : #3217)
• [x] Generate JsonSchema from methods for GPT Function calls (see #3243)
• [x] Automate function calls from GPT function results (see #3243)
• [ ] Access to json documents is a lot of code, need some indexer and / or xpath selection (see #3243)
• [ ] GPT Function does not respect argument schema, array of strings are completed as single string and enum values are ignored
• [ ] Azure OpenAI SDK FunctionDefinition and FunctionCall are hard to use as they partial deserialize the json paylaod and leave the rest as plain strings (FunctionDefinition.Parameters and FunctionCall.Arguments are not usable objects but binary paylaod or string)
• [x] Ability to sample at random from collections (related PRs #3294 )
• [ ] simplify calculations like centroids (related Prs #3298 )

Data preparation issues

• [ ] create float[] column types in dataframe (needed for data prep)

Ecosystem issues

• [ ] Support for easy scraping of wikipedia (maybe using this https://github.com/CXuesong/WikiClientLibrary/tree/6be126af214b60367637a804863b7d7f46f649fe)
• [ ] pdf parsing / extraction
• [ ] arxiv search