An HTML block that starts with `<pre`, `<script`, `<style` or `<textarea`
ends at a line containing `</pre>`, `</script>`, `</style>` or
`</textarea>`, and the spec makes that match case-insensitive (4.6, start
condition 1: "case-insensitive; it need not match the start tag"). The
end tag was found with a case-sensitive `contains`, so `<script>` closed
by `</SCRIPT>` (or `<PrE>` by `</pRe>`) did not end the block: every
following line, up to a lowercase end tag or the end of the document,
became raw HTML, including Markdown after a blank line. cmark and
commonmark.js end the block there.

A block that starts with `<pre`, `<script`, `<style` or `<textarea`
ends at a line containing `</pre>`, `</script>`, `</style>` or
`</textarea>`; the end tag "need not match the start tag" (spec 4.6,
type 1). Only the start tag's own end tag was looked for, so in

    <script>
    x
    </TEXTAREA>
    *y*

the block went on to the end of the document; cmark and commonmark.js
end it at `</TEXTAREA>` and render `<p><em>y</em></p>`.

Look for all four.

Per review: instead of sliding a window over the line once per tag, jump
between the `<`s with memchr and check all four end tags at each one, in a
single pass. On lines typical of script/pre/style blocks this is about 6x
faster than the windowed scan (39 vs 225 ns per line; checking the four
tags in separate memchr passes is 94 ns). The other end conditions (`-->`,
`?>`, `]]>`, `>`) have no letters, so they are matched exactly.