Users should explicitly request cssCoverage in each test rather than
having it run automatically for all tests. The expect re-export was
unnecessary since users import it directly from @playwright/test.

…Tests usage

- Tests cover: configured output directory, default directory, file naming
  from titlePath, and file content being valid JSON coverage data
- Drop scope: 'worker' from cssCoverageDir so it can be set per describe
  block with test.use() — worker-scoped options can only be set at file
  top-level which is too restrictive for users
- Use mergeTests() instead of test.extend() when composing the fixture,
  which is the correct Playwright API for merging test objects
- Update README to reflect mergeTests usage

- Extract slugify() as a named exported function in the playwright fixture
  with a comment explaining why it's needed (titlePath entries contain '/'
  which would otherwise create subdirectories)
- Import slugify in tests instead of duplicating the regex logic
- Extract repeated HTML string to TEST_CONTENT const in tests
- Collapse beforeEach auto-coverage example to a single line in README

Avoids importing @playwright/test entirely, removing the implicit
dependency that would require it as a peerDependency or dependency
in the published package. Users wire save_css_coverage() into their
own Playwright fixture, keeping full control over the fixture lifecycle.

Tests no longer need a browser — mock coverage data is passed directly
to save_css_coverage(), making them faster and simpler.

The subpath added no real value once reduced to a plain helper function.
The full fixture setup is simple enough to paste directly, and keeping it
in the README means users own the code and can adapt it freely.