The bug is with the case of multiple-questions with targeted
feedback within a single problem.
Refactor the targeted-feedback tests to house
the most common xml in external files.
Move the capa test load_fixture function to __init__.py so it
can be used generally.
