AI implementation

We can for example provide JSONs with all the available classes, and sources to include those classes. Then we can provide all fqnames with a JSON describing the classes, like methods and properties.

For example:

## `classes.json`

```
[
{ "class": "korlibs.math.geom.Vector3", "source": "maven:com.soywiz.korlibs.korma:korma:4.0.5a" },
{ "class": "org.jbox2d.collision.AABB", "source": "kproject:https://github.com/korlibs/korge-box2d/tree/0.0.1/korge-box2d##57951cb11dcdccf657d432fa840d6e07532507e8
" },
]
```

## `korlibs/math/geom/Vector3.json`

```
{ "implementations" : ["..."], "parent": "Any", "methods": [...], "properties": [...] }
```

Then we can use OpenAI function calling:

https://platform.openai.com/docs/guides/gpt/function-calling

So everything is resolved serverless with just providing static files heres and published thru pages.

The issue mentions classes.json and korlibs/math/geom/Vector3.json, along with OpenAI function calling and static publication through Pages. Start by clarifying the data schemas and publication approach; done means the class metadata is available as static JSON in a form usable for the proposed function-calling workflow.

Written by the indexing model from the issue text.