FIX: Allow multiple inlined image data links in html clean

[beledouxdenis]

Add a lazy quantifier in the regex _find_image_dataurls to match as few characters as possible,
to make it stop at the first occurence of ;base64,

e.g.

>>> _find_image_dataurls = re.compile(r'data:image/(.+);base64,', re.I).findall
>>> _find_image_dataurls('<div style="background: url(data:image/jpeg;base64,foo); background-image: url(data:image/jpeg;base64,foo);"></div>')
['jpeg;base64,foo); background-image: url(data:image/jpeg']

>>> _find_image_dataurls = re.compile(r'data:image/(.+?);base64,', re.I).findall
>>> _find_image_dataurls('<div style="background: url(data:image/jpeg;base64,foo); background-image: url(data:image/jpeg;base64,foo);"></div>')
['jpeg', 'jpeg']

This allows to have multiple image data links on the same line, which happens for instance in inline styles.

Without this change, _has_javascript_scheme returns True because the count of safe image urls is lower than the number of possible malicious scheme.
Then, the whole style is dropped as considered malicious.

[frenzymadness]

Thank you for your contribution. I'll look at this closely but not sooner than next week.

[frenzymadness]

I like this change. The problem in the CI is unrelated to this PR. I'm gonna try to fix it and then merge the PR. Do you need a new release?

[beledouxdenis]

The problem in the CI is unrelated to this PR. I'm gonna try to fix it and then merge the PR.

No problem, take your time :)

Do you need a new release?

We do not need. We use lxml 4.8.0 and we cannot, in our stable releases, suddenly bump the version to a new release neither we can suddenly add a new dependency to this lxml_html_clean lib, so we are screwed anyway. We will most-likely monkey-patch this locally in our project.

[frenzymadness]

In the end, no hotfix on our side is needed. Thank you once more.

[beledouxdenis]

Thanks 😚