Both filters consume HTML on the input side. Adding `pre_escape => 'html'`
makes the autoescaper escape attacker-controlled inputs before the filters
process them, so they are no longer reachable via `{{ user_input|inline_css }}`
or `{{ user_input|inky_to_html }}` without an explicit `|raw`.
The `html_to_markdown` filter emits plain Markdown text, so the
`is_safe` annotation is dropped entirely and autoescape now handles
its output according to the surrounding context.
The `markdown_to_html` and `inline_css` filters emit HTML, not text
safe in every escaping context, so `is_safe => ['all']` produced
unescaped HTML when their output was interpolated into a JS, CSS or
URL context. The annotation is now `is_safe => ['html']`.