[zayeem06]

Tilde code fences can be closed by the code they contain

When a code block's info string contains a backtick, the commonmark renderer switches to a ~ fence, but it still picks the fence length from the longest run of backticks in the code. So a code block containing a ~~~ line gets written with a three-tilde fence, and that line closes the block early when the output is parsed again; everything after it is read as Markdown, so raw HTML that was inert inside the code block becomes live:

$ printf '~~~~~ a`b\n~~~\n<b>x</b>\n~~~~~\n' | cmark -t commonmark | cmark --unsafe
<pre><code class="language-a`b"></code></pre>
<b>x</b>
<pre><code></code></pre>

The fix makes the helper count runs of whichever fence character is used, so the fence comes out one longer than the longest tilde run. Backtick fences behave as before. I added a case to the render_commonmark api test. It fails without the change, and the full ctest suite (including roundtrip and an ASan/UBSan build) passes with it. Worth checking: I renamed longest_backtick_sequence to longest_char_sequence since it now takes the character; it has only this one caller.