Adopts what the platform's parser now supports: \ instead of the
deprecated \, a stable ID: on every activity, REFERENCE from the
comprehension exercises to their reading passage, and inline gap answers
(____{hay}) so each blank in a multi-gap item is graded on its own.
Also replaces INPUT: either, which the parser silently downgrades to
'speak' — 37 exercises (translation drills, mini-tests, final tasks) were
therefore refusing typed answers. They are now type or speak per exercise,
with LONGFORM on the writing tasks.
Co-authored-by: Cursor <cursoragent@cursor.com>
