Enhance the processing of invalid CSS

[benoit74]

In Zimit2, we are rewriting CSS (both inline and in plain files).

The process is:

• parse the CSS with tinycss2 and rewrite each CSS rules based on a fine understanding of their meaning
• if it fails (probably due to some invalid CSS), rewrite the CSS with simpler regex rules

We might want to push this even further by searching which CSS rule is invalid (tinycss2 is quite tolerant), rewriting all valid ones with the "fine understanding of their meaning" and apply regex rules only to the invalid CSS rules (which have a significant chance to not processed by the browser anyway). This makes sense mostly in the plain file scenario, inline CSS is not expected to contain many rules.

This has already been discussed a bit in #175 where @mgautierfr said:

I have think about that, but not an easy task. We would have to reimplement tinycss2 serialize feature. And on top of that, it is not clear about what to do in case of parsing error. Even tinycss try to be smart and recover error but give up on parsing error (Kozea/tinycss2@main/tinycss2/ast.py#L105-L115).

I consider this is not needed for Zimit2 project completion (at least based on our current understanding of the situation).