perf: faster xml names

Names read ASCII bytes without decoding, attributes are copied
only to sort them, and ignored ranges are looked for only in a
file that has <!-- prettier-ignore-start -->. About 7% less time
on docbook-xsl with one thread.