Create lite-data dataset for JupyterLite

[natinew77-creator]

Problem

The JupyterLite docs build currently downloads the full sample dataset (and a few others) just to serve a small handful of files to the browser notebooks. Those files then get bundled into the documentation build, which bloats it.

Streaming directly from OSF at runtime isn't possible: Pyodide runs in a web worker and OSF doesn't send CORS headers, so the fetch fails from the browser.

Proposal

Create a small curated dataset — lite-data — containing only the files the JupyterLite notebooks actually need (a slim subset of sample). Host it on OSF as its own dataset.

Since OSF supports downloading individual files, we could add something like mne.datasets.lite_data that fetches only the required files rather than pulling the whole sample dataset. Any mne.datasets.* calls in the notebooks could then be pointed at lite-data instead of the full dataset.

This would cut down both the download at build time and what gets bundled with the docs.

Relevant code: mne/datasets/_fetch.py (uses pooch) and mne/datasets/utils/.

Context: PR #13925.

cc @teonbrooks

[palakbhati]

Hi @teonbrooks, I'd like to work on this if it's still open.

My approach: audit the JupyterLite notebooks to identify the exact files from sample (and any other datasets) they actually touch, curate those into a new lite-data OSF project, then add a mne.datasets.lite_data fetcher in mne/datasets/_fetch.py following the existing pooch-based pattern used for other datasets. I'll update the affected notebooks to point at lite_data instead of the full dataset, and check the docs build size before/after to confirm the reduction.

I'll start by going through PR #13925 and the notebooks under the JupyterLite docs build to scope the file list.

[drammock]

@palakbhati this is party of an ongoing GSoC project by @natinew77-creator, so it would be better if you chose an issue not related to jupyterlite