Enable Dask by default for chunked indicators

[akritaag]

Summary

This fixes the Dask issue from my review, following our discussion in #164.

With chunked input, lsa.run() currently raises an error unless Dask options are passed explicitly. The two-line fix defaults chunked indicators to dask="parallelized", while respecting any setting the caller provides. Dask infers the output dtype unless one is supplied, as described in xarray’s API.

The supporting changes add a lsapy[dask] install option, a chunked example, docstrings, a changelog entry, and tests for lazy execution, numerical agreement, and explicit overrides. The example also runs in a fresh Python process.

Validation

Tested on Python 3.12.14, macOS arm64, with xarray 2026.7.0, Dask 2026.8.0 and NumPy 2.5.3.

• Full suite: 100 passed, including 11 Dask and example cases. The existing 7,507 warnings remain. The test runner redirected Pooch’s cache to a writable local directory.
• Applicable pre-commit hooks passed for the changed files.
• The chunked documentation example runs successfully.
• Strict Sphinx HTML build reports four existing unknown-link-target errors for xarray, Dask and matplotlib in docs/index.rst, which is unchanged from the base branch. Existing notebooks were rendered without execution.

No performance benchmark or cross-platform test matrix was run locally.

Pull request checklist

• Addresses an existing issue: LSAPy Review #2 #164 (Dask item)
• Tests added
• Documentation updated
• CHANGELOG.rst updated

AI assistance

OpenAI Codex was used in an agentic workflow to prepare the two-line runtime change. The supporting changes are mostly tests and documentation, plus contributor credit in AUTHORS.rst and .zenodo.json. Automated checks do not establish human review or pyOpenSci acceptance.

[baptistehamon]

Thanks a lot @akritaag! I just suggested minor changes mainly for consistency with other files. Feel free to also update the the AUTHORS.rst and .zenodo.json if you want!

[akritaag]

Thanks @baptistehamon! Applied the naming and wording suggestions in e4e77ec, and added my name under Contributors in AUTHORS.rst and .zenodo.json. Thanks for suggesting it!

I made a small adjustment to the override assertion and analysis assertion: both compare against exp. That preserves the comparison with the unchunked result; res also isn't defined in the analysis test.

On the example test, the new example is an .rst file, so nbval doesn't collect it. Running pytest --nbval --collect-only -q docs/notebooks/ collected 12 cells from the existing notebooks, but not this example. I've kept the test and added a docstring explaining why. The nbval documentation describes its notebook collection.

All 11 Dask/example tests and Ruff checks passed locally. Read the Docs and pre-commit.ci are now green too.

[baptistehamon]

LGTM ! Thanks again @akritaag !