Generate SVG datasets

[MischaPanch]

Using existing/commercial model providers may prove to be a limiting factor. If we require models to gain a deeper understanding of SVGs and, more generally, vector graphics, then we may need to train them using curated datasets.

Note that providing datasets may generate visibility in the community.

We could leverage existing dataset annotation systems, inclusion of visual renders and other techniques.

There are multiple ways to go about this in all aspects: from size (small specialized DS for in-context learning, fine-tuning dataset, preference datasets for "alignment") to content (refactoring, generations, translations into different formats), to the generation method (completely unsupervised and synthetic to based on actual user data, and at least partially supervised).