Release phylib 2.7.1

[rossant]

Summary

Prepare phylib 2.7.1 for release, including the template-less waveform and spike-selection fixes required by cortex-lab/phy#1404.

• Document all changes since 2.7.0, including atomic curation writes, large-dataset optimizations, geometry and loading fixes, and development improvements.
• Set the package version to 2.7.1 and date the changelog for the 8 October 2026 release.
• Fix existing flake8 violations and enforce lint in CI.
• Include Ensure dataset is writable #66's test-fixture permission fix: copied datasets remain writable when packaging systems provide a read-only source cache. This changes test preparation only.
• Correct the wheel tag to Python 3 only and include the changelog and test dependency list in source distributions.
• Replace the obsolete README release commands with isolated builds, artifact checks, clean installation verification, and downstream validation.

Validation — 2 October 2026

Validated locally with Python 3.12.14:

• flake8 7.4.1: passed.
• Complete phylib suite: 280 passed.
• uv build: wheel and sdist built; both passed twine check.
• Archive inspection: source includes changelog, test dependencies, fixtures, and probe data; wheel metadata reports 2.7.1 and py3-none-any.
• Clean wheel installation: package version and imports verified from the installed wheel, without an editable phylib checkout.
• Downstream phy #1404 regression test_template_controller_without_templates_uses_stored_waveform_channels: passed against that wheel, checking stored waveform sample count, per-cluster channel rankings, and distinct Probe View positions.

All seven GitHub CI jobs passed on the refreshed PR head 363af39: lint, full Linux tests on Python 3.10–3.12, full macOS tests on Python 3.12, Windows smoke tests, and package build.

Release checklist

• Changelog includes all changes since 2.7.0, including the October test-fixture fix
• Package version is 2.7.1
• Local flake8 passes
• Local test suite passes
• Wheel and sdist pass twine check
• Built wheel passes a clean installation and import smoke test
• Downstream phy template-less waveform/channel regression passes against the built wheel
• GitHub CI passes on the final PR head
• Approved by @oliche
• Changelog receives the actual release date (8 October 2026)
• PR merged and merged master CI verified
• Annotated 2.7.1 tag created from the merged release commit
• Release artifacts rebuilt and checked from the tagged commit
• Tagged wheel installation and downstream regression verified
• Artifacts uploaded to PyPI
• PyPI installation verified
• master bumped to 2.7.2.dev0 with an Unreleased changelog section (Start phylib 2.7.2 development after the 2.7.1 release #67; post-merge CI passed)

@oliche approved the release on 6 October 2026. Cyrille authorized merging and publishing on 8 October 2026; execution is proceeding with fresh CI and tagged-artifact verification.

Publication — 8 October 2026

Merged as 457915b, tagged 2.7.1. All seven release and merged-master CI jobs passed. Tagged-source lint and all 280 tests passed; wheel and sdist passed archive/metadata checks and twine check. The tagged wheel passed a clean installation and the downstream template-less waveform/channel regression. Published PyPI files have SHA-256 hashes identical to the tagged artifacts; exact-version installation from PyPI passed. GitHub release assets are the same verified files. The post-release development bump was merged in #67 as ca1e063; all seven post-merge CI jobs passed.

[rossant]

Release preparation is complete and the full GitHub CI matrix is green, including lint and package build. The PR is now paused at the required @oliche approval gate; it will not be merged, tagged, or published before explicit approval.

[oliche]

All good, the ibl-sorter integration tests pass.

Ok to merge on my end !