test: generate_alphas/organize_clips/scan_clips coverage, opencv guard, DeepWiki link

[shezmic]

What changed

• tests/test_clip_manager.py — adds 13 new tests across three new classes (closes Testing: Add coverage for run_videomama, generate_alphas, organize_clips, scan_clips #19):
  • TestGenerateAlphas (4 tests): verifies GVM is skipped when alpha already exists, triggers process_sequence for missing-alpha clips, handles ImportError gracefully, and confirms generated files are renamed to match input stems.
  • TestOrganizeClips (5 tests): missing-dir warning, loose-video move + hint-dir creation, IgnoredClips/Output skip, and subdir dispatch.
  • TestScanClips (4 tests): CLIPS_DIR auto-creation, valid clip discovery, frame-count mismatch exclusion, and hidden/underscore dir filtering.
• pyproject.toml — adds [tool.uv] constraint-dependencies guard so any future transitive dep that pulls in opencv-python-headless fails resolution explicitly instead of silently corrupting the environment (closes Add uv override to prevent opencv-python / opencv-python-headless conflict #17).
• README.md — adds DeepWiki link in the Advanced Usage section (closes Add DeepWiki link in Readme #85).

Why

• Testing: Add coverage for run_videomama, generate_alphas, organize_clips, scan_clips #19: generate_alphas, organize_clips, and scan_clips had zero test coverage despite being core pipeline orchestration logic.
• Add uv override to prevent opencv-python / opencv-python-headless conflict #17: The current uv.lock resolves cleanly, but is fragile — a future diffusers/imageio/PIMS release could silently pull in opencv-python-headless, which collides with opencv-python at the file level.
• Add DeepWiki link in Readme #85: DeepWiki provides auto-generated, navigable codebase docs useful for contributors.

How to verify

uv sync --group dev
uv run pytest tests/test_clip_manager.py -v   # 50 passed
uv run pytest -q                               # 208 passed, 2 skipped
uv run ruff check tests/test_clip_manager.py pyproject.toml

🤖 Generated with Claude Code

[Mogi83]

Hey! I see you've bundled #19, #17, and #85 here. I actually have a focused PR for Issue #19, being PR #111 but will look over the tests made for #19 coverage. To keep the git history clean and simplify debugging if a regression occurs, let's keep these changes in individual, atomic PRs ideally.

As far as #17 goes an attempt has already been made in PR #62 if you can explain the AI code it will be considered!

[shezmic]

Thanks for the detailed feedback @Mogi83 — completely agree on atomic PRs. I'll close this and open separate ones.

On #19: I see PR #111 already provides comprehensive coverage (32 tests, 51% clip_manager.py coverage). Happy to defer fully — if any of the 13 tests here are useful to pull in, feel free; otherwise I'll drop them.

On #17 vs PR #62: The difference between the two approaches:

• PR Fix #17: Override opencv-python-headless #62 uses override-dependencies with extra == 'never' — a clever marker trick that makes the package conditional on an extra that's never activated, effectively silencing it from the resolver without surfacing an error.

• This PR uses constraint-dependencies = ["opencv-python-headless==99999"] — adds a hard constraint at resolution time. If any current or future transitive dep requests opencv-python-headless, uv fails explicitly with a conflict message rather than silently ignoring it. Version 99999 is a sentinel that doesn't exist on PyPI, so resolution always fails loudly.

The practical distinction: constraint-dependencies is uv's documented mechanism for "this package must never be installed" and produces an actionable error. The extra == 'never' trick is semantically opaque to future maintainers and silently swallows the conflict instead of surfacing it.

I'll open a standalone atomic PR for #17 only so it can be evaluated on its own merits alongside PR #62.

[shezmic]

Closing in favour of atomic PRs. Will open separate PRs for #85 and #17 individually.