Feat: add dataset module in tracel core

[Zatiji]

This branch adds dataset streaming from Station, so you can pull a Station-hosted dataset into training as a burn::data::dataset::Dataset<T>.

Concretely:

• A DatasetProvider trait + DatasetModule (tracel-core/src/dataset/mod.rs): the abstraction for fetching pages of raw dataset items over the network.

• AnnotationDataset<T> (burn.rs): wraps that in burn's Dataset<T> trait: builds a lightweight index of valid items on first use, decodes each item as JSON into T, skips malformed items cleanly.

• Station wiring (station.rs): implements DatasetProvider by calling tracel-client's dataset HTTP endpoints, plus a not-found error path that reports whether it's the dataset or the version that's missing.

• Plugged into Context: context.datasets() returns a DatasetModule when connected via Station, None otherwise, same pattern as the existing model_registry().

• Exposed publicly as tracel::dataset::{AnnotationDataset, DatasetModule}.

Usage: context.datasets().unwrap().as_burn_dataset::<MyItem>(name, version) gives you a burn::Dataset<MyItem> ready to hand to a Dataloader.

Note: caching is out of scope for this PR. It'll be added in a follow-up once the core logic here is approved

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.