ie-html: implement CR/CRLF normalization in tokenizer

[thomasnemer]

Goal

The WHATWG spec requires preprocessing that normalizes \r\n → \n and lone \r → \n before tokenization.

Current state

11 html5lib tokenizer tests fail due to missing CR normalization:

• CR in comments, bogus comments
• CRLF sequences not normalized

Implementation

• Add preprocessing pass in Tokenizer::new() or handle in next_char()
• Replace \r\n with \n, lone \r with \n
• Or handle during character consumption

Impact

Fixes 11 failing conformance tests, improves pass rate from 99.6% to ~100%.