fix(skins): replace placeholder artwork

[luinbytes]

Why

The Neon Cat and Retro Cat selectors loaded the right skin IDs, but every bundled PNG was an exact copy of the default artwork. Users could save a skin choice without seeing a visual change.

Scope

• Replace all six Neon Cat and Retro Cat animation frames with distinct 200 by 126 transparent artwork.
• Add a regression test that rejects duplicate built-in frames and dimension drift.
• Remove stale placeholder notes from both skin README files.
• Set the package version to 2.0.9.

The Settings Apply callback, config persistence, skin discovery, and Qt lifecycle code are unchanged.

Blast Radius

This changes the bundled Neon Cat and Retro Cat appearance in the desktop app and tray icon. Default Cat stays unchanged. Matching dimensions preserve the current window and animation geometry.

Verification

• python3 -m unittest -q tests.test_skin_manager.TestBuiltInSkinAssets.test_builtin_skin_assets_are_distinct_and_keep_dimensions passes.
• Direct PNG inspection confirms that all six frames are 200 by 126 RGBA files with transparency.
• The full local suite reaches 19 passing tests and four skips, then fails to import test_config because this host lacks PyQt5. GitHub Actions installs the project requirements under Python 3.11 and runs the complete suite before packaging.
• git diff --check passes.

Closes #10

[luinbytes]

Independent verification verdict is PASS+NOTES.

I checked PR #15 at head 1d1d9969677a887c6e2af95bb7040903e10f9769 against base 32b31134561f2a3a8eef5954217b023e8096f0d3. The stable patch-id is 211de18d780d835ba8c1ec05d54a43c546937d48.

• The focused built-in skin asset test passed.
• All nine built-in pose files are valid 200x126 PNGs with transparent pixels. Neon and retro differ from default. Each skin has distinct rest, left, and right frames on the same canvas.
• Release run 33829786403 passed on macOS, Ubuntu, and Windows. The Windows artifact is a PE32+ x86-64 executable. Its PyInstaller archive contains all six changed neon and retro files, with embedded hashes matching the source files.
• The local Python 3.14 environment lacks PyQt5, so the complete suite could not run locally. CI ran the full suite on Python 3.11 with the workflow's headless settings.
• This manual workflow run built artifacts but skipped GitHub Release publication. A tag run is still needed to publish v2.0.9.

The patch is ready for release publication and reporter retesting.