This restores functionality that has been broken since the introduction of
XModuleDescriptor/XModule proxying (part of the XBlock transition). It generates
a CSV of all answers for all content of type "problem" in a given course, with a
row per (problem part, answer). The format is:

url_name, display name, answer id, answer, count

Example values:
  url_name = "7f1b1523a55848cd9f5c93eb8cbabcf7"
  display name = "Problem 1: Something Hard"
  answer id = i4x-JediAcdmy-LTSB304-problem-7f1b1523a55848cd9f5c93eb8cbabcf7_2_1
  answer = "Use the Force"
  count = 1138

Since it only grabs things of type "problem", it will not return results for
things like self/peer-assessments. Any Loncapa problem types will show up (so
multiple choice, text input, numeric, etc.)

Instead of crawling the course tree and instantiating the appropriate CapaModule
objects to grab state, this version draws directly from StudentModule. This lets
us skip a lot of processing and most importantly lets us generate the answer
distribution without causing side-effects (since XBlocks auto-save state). It
also lets us take advantage of a read-replica database if one is available, to
minimize locking concerns.

There are minor changes to the legacy dashboard around CSV charset encoding and
a change to OptionResponseXMLFactory to make it more unicode friendly. Answer
distribution output is now also sorted, to group together answers for the same
content piece.

Note that this does not introduce celery into the process. Answer distributions
are still only available for small courses.

This was originally created to fix [LMS-922], but it also addresses [LMS-811] and
possibly other areas in the legacy dashboard where CSV downloads break due to
character encoding issues.