HTML::Node.parse shouldn't blow up on unclosed CDATA sections

Currently, the HTML scanner's Node.parse will raise an exception if it hits an unclosed CDATA section.

With this patch, if an unclosed CDATA section is encountered and parsing is strict, an exception will be raised. Otherwise, we consider the remainder of the line to be the section contents. This is consistent with how CDATA sections are handled by the tokenizer.

This patch also adds sanitization tests for valid and unclosed CDATA sections.