It appears that for random text the MIME lexer wins:

    In [4]: import pygments.lexer
    In [5]: pygments.lexers.guess_lexer("Hallo")
    Out[5]: <pygments.lexers.MIMELexer>

which usually generates errors for random text:

    In [11]: from markdown.extensions.codehilite import CodeHilite
    In [12]: CodeHilite(src = 'Hallo').hilite()
    Out[12]: '<div class="codehilite"><pre><span></span><span class="err">Hallo</span>\n</pre></div>\n'

which just look ugly (red border on each line).

So switch off language guessing. In that case, code blocks without a
explicit language are left as is:

    In [13]: CodeHilite(src = 'Hallo',guess_lang=False).hilite()
    Out[13]: '<div class="codehilite"><pre><span></span>Hallo\n</pre></div>\n'

In Markdown, the language should be set explicitly, for language
specific highlighting.

    :::python
    print("Hallo")

The bulma style sets margin-bottom to 1em for several block-level
elements (eg <p>). However the codehilite <div>s, generated by the
codehilite Markdown extension don't have that. So we have 1em vertical
space before a code block but not after, which just looks wrong.

Add vertical space after codehilite <div>s.

Add Admonition extension which allows to add class attributes and a
title to a block. So

    !!!alert "Red Alert"
        Warp core breach

becomes

    <div class="admonition alert">
      <p class="admonition-title">Red Alert</p>
      <p>Warp core breach</p>
    </div>

Add inline style to wagtailadmin/pages/edit.html to force the markdown
textarea to monospaced font.