Update the cached dataset path shown in the quickstart

[mvinyard]

Follow-up to scDiffEq/scDiffEq#113.

What changed upstream

scDiffEq now caches the raw download and the preprocessed result as separate files, so sdq.datasets.larry() loads from larry.processed.h5ad rather than larry.h5ad. Previously both shared one filename, which meant a dataset already on disk was returned unpreprocessed with no X_pca.

Why it matters here

tutorials/quickstart.ipynb renders on scdiffeq.com, and its stored output still showed the old path:

scDiffEq [INFO]: Loading data from /Users/mvinyard/data/scdiffeq_data/larry/larry.h5ad

A reader following that would go looking for a file the loader no longer writes. (Existing caches are auto-migrated, so the old name disappears on first use.)

Note on how this was fixed

This edits the recorded output rather than re-executing the notebook: that cell downloads 5.3 GB and the notebook goes on to train a model, so a full re-run was not practical.

The single replaced line is exactly what the loader emits today on a cached run, and nothing else changed — the AnnData repr (130,887 × 2,492) still matches the default variant, and the scaler.pkl / pca.pkl paths further down remain correct, since the default variant deliberately keeps those bare filenames.

Diff is one line; nbformat.validate passes.

Not touched

manuscript/tmp_revision/figure_r1/notebooks/FigureS9.ipynb shows the same old path. Left alone deliberately — manuscript notebooks record what was actually executed for the paper, and that directory is not published to the docs site (the workflow copies manuscript/figure_*/notebooks/, not manuscript/*/figure_*/).

🤖 Generated with Claude Code

[mvinyard]

Added a second commit removing .github/workflows/docs.yml, which is what was failing on this PR.

It turned out to be pre-existing, not caused by the notebook change, the same workflow has been failing on every push to main:

failure  pull_request  fix/quickstart-cache-path  Build and Deploy Docs
failure  push          main                       Build and Deploy Docs
failure  push          main                       Build and Deploy Docs

The file is a verbatim copy of the scDiffEq repository's docs workflow — its first step is even named Checkout scDiffEq, but running here it checks out this repository, which has no docs/ directory. It then clones this repository again, copies notebooks into a docs/source/ it just created, and dies at:

ERROR: Could not open requirements file: 'docs/requirements.txt'

It cannot succeed here, and if it did build it would deploy to this repository's GitHub Pages rather than scdiffeq.com. The site is built from the scDiffEq repository, whose own workflow already clones this one for manuscript/figure_*/notebooks/ and tutorials/.

One consequence worth knowing: notebook changes here don't rebuild the site on their own: that happens on the next scDiffEq build. Closing that gap properly means triggering the scDiffEq workflow from here via repository_dispatch, which needs a token with workflow scope on that repository. Deliberately not added here, since it's a credential decision.