E2E tests comparing preset output against Tailwind CSS v4

The `zoecss-presets-tailwindcss4` preset should produce CSS output identical (or intentionally divergent) to actual Tailwind CSS v4 for the same input tokens. E2E tests would run both engines on a shared set of utility classes and diff the results, catching regressions and undocumented divergences.

Shares fixture needs with #12 (realistic inputs), but orthogonal: #12 focuses on performance benchmarks, this focuses on correctness verification. Not in scope: performance measurement (covered by #12), testing non-Tailwind presets.

## Current state

Tests are unit-level per rule/module. No pipeline-level comparison against Tailwind CSS v4's actual output exists.

Start by tracing the existing unit-test pipeline and the realistic fixture work referenced in #12, then determine how the zoecss-presets-tailwindcss4 preset and Tailwind CSS v4 can run on shared utility classes. Done means E2E tests compare both outputs, record intentional divergences, and exclude performance and non-Tailwind presets.

Written by the indexing model from the issue text.