Fix answer fetching

[satyamtg]

This refactors the problem answer fetching and uses a proper POST request to the API for getting the answers as follows -

1. generate all combinations of possible answers
2. for every candidate answer, do the POST request to get the HTML content that which is to be put (this contains all messages, and classes for styling)
3. Store the content in a JS file for each problem as follows -

   problem_answers[problem_id] = {<answer_id>: <html_content>, ... , <answer_id>: <html_content>}

4. Use this problem_answers object to replace the content of the HTML according to the IDs of selected answers.

The point to note here is this is almost the same process as what's done in the online version.

It contains the following changes -

• update_csrf_token_in_headers() is introduced in instance_connection.py to update the latest CSRF token before making any request to the instance server. The previous method of setting CSRF token at the start has been dropped.
• The problem template is fully refactored to contain the original HTML from the instance as far as possible with minimum changes.
• In the vertical template, the workaround to emulate the sequential content change behaviour is refactored and can now be called from anywhere. This allows calling it after the problem HTML change to re-initialize any action listeners.
• problem.py is fully refactored as follows -
  • clean up the checked classes
  • clean up the messages from problem checking (if problem was previously checked on the instance)
  • get_answers() is refactored to get answers using the same POST requests as the online version
  • check_problem_type_and_get_options() is added. It returns whether answers are fetchable for a problem, if yes, also returns whether the problem has single correct answers, and a list of options for answers.
  • check_answer() is added to do the POST request with a combination of answers
  • write_answers_and_mark_available() is added to write the answer content in a JS file in instance_assets directory
  • render() now passes only the required content to jinja
  • autogenerated HTML tags are now cleaned up
• xblocks.js is fully changed. All previous functions are dropped. A new function problem_check_answers() with the same name has been added to get the selected answer ID and accordingly change HTML content.
• Answer showing is fully dropped due to the following reasons -
  • The default behaviour on instances is to enable this button only if the student answers correctly or has already answered correctly previously
  • We don't store any such usage data in the ZIM
• README.md is updated to clearly mention the limitation in answer fetching
• CHANGELOG.md is updated

This fixes the following issues - #85, #35 Also, I wanted to look into #27, but this says that edX doesn't support Adaptive hints. Also couldn't find any adaptive hint MOOC. @dattaz could you clarify on this? I think we may want to drop this if there's no MOOC with this feature.

[rgaudin]

@satyamtg ?

[satyamtg]

Oh. Sorry @rgaudin , I forgot this. For dropdowns, at this point of time, I didn't implement those. However, that's doable for dropdown questions having a single dropdown or group of dropdowns having total sum of all options in all dropdowns <= 5. But wherever dropdowns are present, they're actually multiple dropdowns in a single question and each dropdown has several options as mentioned in https://github.com/openzim/openedx/issues/35#issuecomment-671872999 . What do you suggest?