fix: catastrophic backtracking in Core.AggressivelyFixLt

[bytestream]

When provided with a large HTML document (over a million characters) the Core.AggressivelyFixLt regex results in catastrophic backtracking and $html = null being returned. TLDR; HTMLPurifier gives you back a null document...

I tried many times to produce a regex which did not suffer from catastrophic backtracking but I think it ultimately comes back to the argument of why you should not use regex to parse HTML. The only solutions I could come up with were to either:

• Increase pcre.backtrack_limit to a higher value
• Disable Core.AggressivelyFixLt but that's sub-optimal given the approach seems to work on documents of a reasonable size...
• Handle the null return value from preg_replace_callback and return $html (disable armor logic if a regex error occurs)

The solution in this PR uses a little algorithm which employs only standard string manipulation functions so it works incredibly fast. The algorithm searches for HTML comments and allows a callback to be ran on them.

I've not messed with the signatures of the callbackUndoCommentSubst and callbackArmorCommentEntities functions because they're public and might be used by other libraries.