Release/v0.2.0

[degoldschmidt]

Cuts v0.2.0 — the first tagged release the project will have had — and clears two
things that were quietly broken underneath it.

The user-facing changes themselves are already on main (faceting, QC auto-removal, arena
alignment, and the fixes for #1 and #2); see CHANGELOG.md for the full entry. This PR is
the release itself plus the packaging, docs and CI work that turned out to be prerequisites
for one.

Release

Version 0.1.0 → 0.2.0 in pyproject.toml and uv.lock, the [Unreleased] changelog
section dated and closed, README status line refreshed.

The sdist was unpublishable

uv build was producing a 124 MB source distribution: data/sample holds ~340 MB of
raw recordings and hatchling was sweeping all of it in. PyPI rejects any file over 100 MB,
so the release workflow's publish job could never have succeeded — this would have been
discovered only after tagging.

[tool.hatch.build.targets.sdist] now excludes /data (plus /dist, /site), taking the
sdist to 269 KB with all 77 source, test, config and doc files intact. The tests that
need the fixture already skipif when it is absent, so a build from sdist just skips them.

Related, not addressed here: data/sample/20240215/ and data/sample/raw/ hold
byte-identical copies of both recordings (verified by SHA-256) — ~170 MB of duplication in
the repository. Worth consolidating, but it means rewriting history, so it does not belong
in a release PR.

The GUI suite has never run on a runner

tests/test_gui.py importorskips qtpy, and CI syncs without the optional gui group, so
all 14 tests skipped on every push since M8. That is precisely the path #1 lived in: the
GUI worker dropped the provenance step and nothing caught it.

Adds a second gui job that installs the extra plus the Qt runtime libraries the runner
image lacks, then type-checks and runs the suite offscreen. It is a separate job rather
than --group gui on the existing sync, because the no-extra sync is the only thing proving
the package still imports and type-checks when Qt is absent — a property that matters
exactly because flypad.gui.* carries mypy relaxations for that case. The gui job also
re-runs mypy, which with PySide6 present checks the GUI against real Qt signatures instead
of Any.

Both jobs assert their own premise, because both fail silently otherwise: test fails if
qtpy ever appears in the default sync, and gui fails if qtpy is missing after the sync —
without which a broken install would look like a green job that skipped everything.

Docs

alignment and the QC auto-removal toggles had shipped without ever reaching the docs site,
which publishes to Pages on every push to main. docs/configuration.md gains a section
for each, including the warning that moving a dataset from matlab_compat to corrected
changes per-condition counts because the removal takes effect — with per_fly naming the
channels that went.

Verification

• CI on this branch is green on the first run: test (3.11) ✅, test (3.12) ✅, gui ✅ —
  14 GUI tests passed on the runner, with Assert Qt is importable confirming the extra
  installed and the apt set sufficient.
• 246 tests pass locally with the gui extra; 232 + 1 skipped in a runner-faithful
  environment.
• uv build clean → wheel 114 KB, sdist 269 KB. The wheel installs into a fresh venv,
  flypad version reports 0.2.0, the console script runs, and py.typed plus both config
  presets ship.
• uv sync --frozen verified at the release commit specifically (via a throwaway worktree),
  since a lockfile lagging its own version bump would break --frozen for anyone checking
  out the tag.
• mkdocs build --strict clean.

After merging

Tag v0.2.0 at the merge commit on main, not at the release commit on this branch, so
the tagged tree includes the docs and CI work. Pushing the tag fires the release workflow
(build → smoke → publish); publish needs a PyPI trusted publisher and a pypi
environment configured for this repo, and will fail red without them. The name flypad is
currently unregistered on PyPI.