Improve parsing of course archive

[michaellzc]

The current implementation has the following disadvantage

• it assumes each component live in its own directory, learn more
• R/W operations against course XML is not consistent, sometimes it is done on disk directly and it is done in memory. This could potentially lead to inconsistencies between the content on disk and in-memory

How to improve?

Implement some sort of data structure such that we can hold the entire course structure (XML and policies, excluding special assets, such as HTML and static assets).

TODO

• Design the data structure to hold, XML, policies.
• Implement a parser/deserializer to take a course archive as input (potentially empty input where it can be used to initialize an empty course structure)
• Implement a serializer to transform the course structure into a course archive
• Refactor codebases to make all operations against course structure are done in memory. We should avoid introducing breaking change to public API (methods without a _ prefix)