Modernize packaging + CI (setup.cfg -> pyproject, uv-stub CI) and fix two latent threading bugs

[thorwhalen]

Part of the fleet-wide wads repo-modernization pass. This repo was still on the
legacy generation (setup.cfg + setup.py + a hand-rolled ~90-line workflow using
actions/setup-python@v2); it now matches the current standard: a hatchling
pyproject.toml as the single source of truth plus a 5-line ci.yml stub that
calls the reusable workflow i2mint/wads/.github/workflows/uv-ci.yml@master.

⚠️ Merging this PR will publish a new release to PyPI. Pushing to master
runs the publish job, which bumps the version from the declared 0.1.3 and
uploads. 0.1.3 is deliberately the version currently on PyPI, so the automatic
bump lands on the correct next patch rather than re-publishing a version behind
the released one.

1. Packaging: setup.cfg -> pyproject.toml

Full inventory of what was in setup.cfg and where it now lives:

setup.cfg	pyproject.toml
name = pchealthstream2py	[project].name
version = 0.1.3	[project].version
description	[project].description
long_description = file:README.md + long_description_content_type	[project].readme = "README.md"
url	[project.urls].Homepage
license = apache-2.0	[project].license = "Apache-2.0" (SPDX string form; LICENSE file kept, and it is a real Apache-2.0 text)
install_requires = psutil, stream2py	[project].dependencies
packages = find: / include_package_data = True	[tool.hatch.build.targets.wheel].packages
platforms = any	Operating System :: OS Independent classifier
zip_safe, root_url, display_name, description_file	no pyproject equivalent (setuptools-only / wads-internal); dropped intentionally

There were no entry points / console scripts, no package data files, no
extras, no MANIFEST.in and no requirements.txt — nothing else to carry over.
Verified against real imports with wads-deps: the only module-level
third-party imports in the package are psutil and stream2py, which is
exactly what is declared.

Added while here: requires-python = ">=3.10", keywords, classifiers, authors,
Repository / Documentation URLs, the ecosystem-standard [tool.ruff] block
(without it the repo falls through to ruff's moving default and goes red on
unrelated style drift), and the standard .editorconfig.

Manual fixes to the three known rough edges of wads-migrate setup-to-pyproject:
the deprecated [project.license] text = ... table -> SPDX string; the empty
[tool.wads.ci].project_name -> "pchealthstream2py" (it is the target of CI's
ruff-check and coverage steps); and testpaths — see next section.

testpaths — this one actually mattered

wads CI runs pytest --doctest-modules with no path argument, so collection is
driven entirely by testpaths. The migration tool writes ["tests"]
unconditionally, and this repo has no top-level tests/ dir — its tests live in
pchealthstream2py/tests/. Left as generated, CI would have collected nothing
and still reported green. Set to ["pchealthstream2py"], which also matches what
the old workflow ran (pytest --doctest-modules -v $PROJECT_NAME).

Verified in the branch CI run: collected 3 items, 3 passed.

2. CI: legacy workflow -> uv stub

wads-migrate ci-to-uv then wads-migrate ci-to-stub. The old workflow
referenced secrets.PYPI_USERNAME and secrets.PYPI_PASSWORD; the uv CI uses
token-only PyPI auth, so only PYPI_PASSWORD is passed through by the stub and
PYPI_USERNAME is no longer needed. No other secrets were referenced, so
[tool.wads.ci.env] stays empty.

3. Two latent threading bugs fixed (separate commit)

Both are reproducible on the released code and were surfaced while validating
the migration.

a. The stop flag shadowed threading.Thread._stop. StatusInfoReader
subclasses threading.Thread and declared _stop: threading.Event. But
Thread._stop is a method CPython calls internally from
_wait_for_tstate_lock(), which is reached from both join() and is_alive().
On the released version:

reader.join(timeout=2)
TypeError: 'Event' object is not callable

and is_alive() could never flip to False after the worker finished, because
the interpreter's own bookkeeping call raised. Renamed to _stop_event.

b. All readers shared one queue and one stop flag. _index, _data,
_stop and _bt were class attributes, so every StatusInfoReader instance
shared a single SyncQueue and a single Event: opening a second reader
cleared the first one's buffer, closing either stopped both, and both readers'
items interleaved in one queue. Moved to per-instance state in __init__.

Also corrected network_download_speed's return annotation: dict or None
evaluates to plain dict at class-creation time, so the "or None" was silently
discarded. Now Optional[dict], which is what the body actually returns.

Two focused regression tests were added alongside the fixes (the only new tests
in this PR — no test-coverage sweep was attempted here).

Deliberately left undone

• StatusInfoReader still cannot be re-opened after close() — a
  threading.Thread can only be started once, while SourceReader's documented
  contract (its own class docstring) allows close-then-reopen. Fixing it means
  having a thread rather than being one, which changes
  isinstance(reader, threading.Thread), so it is out of scope for a packaging
  pass. Filed as StatusInfoReader cannot be re-opened after close() (SourceReader contract violation) #1 with a repro and a suggested fix.
• No ruff format sweep. The code is still single-quoted; CI's publish job
  formats and pushes back on merge, as it does fleet-wide. Keeping it out here
  keeps this diff reviewable.
• No docs, type annotations, py.typed or agent-skills work — separate
  dimensions of the rollout.

Verification

	Result
Local gate, baseline (before any change)	pchealthstream2py pass 4.1s — 1 suite: 1 pass, 0 fail
Local gate, final	pchealthstream2py pass 4.4s — 1 suite: 1 pass, 0 fail
Local pytest pchealthstream2py --doctest-modules -q	3 passed
Local ruff check pchealthstream2py (0.16.0)	All checks passed
Local uv build	sdist + wheel build clean; wheel contains the package and LICENSE
Branch CI run	https://github.com/i2mint/pchealthstream2py/actions/runs/30860120932 — success

Branch CI jobs: Read Configuration ✅, Validation (3.10) ✅, Validation (3.12) ✅,
Windows Tests ✅, Publish skipped (branch is not the default branch), Publish
GitHub Pages skipped.

Repo audit (python -m wads.repo_audit) before -> after: the HIGH no pyproject.toml, MEDIUM legacy packaging files present and MEDIUM CI is the legacy generation findings all clear, as does LOW missing .editorconfig.
ci_format goes legacy -> uv-stub, legacy_files goes to empty.

The only new finding is LOW GitHub topics (none) != slugified pyproject keywords — a direct consequence of this PR filling in the previously-empty
keywords. The repo's GitHub topics and homepage have been aligned to match
(repo settings, not part of this diff). The remaining LOW findings (no agent skills) are other dimensions of the rollout, untouched here.