cancel() was updating the dirty state in order to determine if it should

show a prompt. However, unlike save(), we don't want to unconditionally

do this, as we can end up in a state where the initial and current

fields don't match, but the user hasn't actually changed anything

(depending on the initial value set in an edited field).
We now call dirty() instead, which will properly handle forcing a

computation only if necessary.

Unit tests pass.
No longer saw a prompt on cancel for Markdown fields.