Add a benchmark

[oliverklee]

To work on improving performance, we need benchmarks to check whether a change helps.

I propose we benchmark these things:

• the same CssInliner instance with the same HTML and CSS
• the same HTML and CSS with a new CssInliner instance each time
• the same CSS with different HTML with a new instance each time (this probably is the real-world use case)
• different HTML, different CSS, separate instances