Harden upscaler CLI and add test coverage

[kernastra]

What changed

• prevent custom output names from escaping the configured output directory
• keep tiny scaled images at valid nonzero dimensions
• handle nested input/output directories, .tif files, EXIF orientation, and JPEG color modes
• add --yes for automation and return nonzero status for processing failures
• remove unused OpenCV and NumPy dependencies
• update installation and usage documentation
• add an eight-test unit/CLI suite and a Python 3.10–3.14 GitHub Actions matrix

Why

The audit reproduced a path traversal issue in custom naming, zero-dimension failures for tiny images, and avoidable installation weight from unused dependencies. The CLI also lacked automated regression coverage and reliable exit codes for scripted use.

Impact

PixelNova is safer for custom naming, more reliable across supported image formats and edge cases, lighter to install, and easier to automate and maintain.

Validation

• python3 -m unittest discover -v — 8 tests pass
• python3 -m compileall -q upscaler.py demo.py example.py setup.py tests
• bash -n run.sh
• git diff --check
• GitHub Actions YAML parsed successfully