Tokenizer: Do not suppress 'missing semicolon after character reference' errors in attributes

[squeek502]

This behavior was explicitly 'for historical reasons,' and because the tokenizer is not actually concerned with whether or not to convert the character reference to the corresponding codepoint(s), there's no downside to diverging from the spec here.

Follow up to https://github.com/kristoff-it/super-html/pull/11#issuecomment-2211764652

[kristoff-it]

❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️ ❤️