Fixed bug in anonymisation function

See merge request !2

The function `do_anonymous` in `hektor.py` was dependant upon the iteration order of a dictionary being the same as the insertion order. This is an implementation detail of CPython3.6 and should not be relied upon.

Changed output format of anonymization mapping to csv

See merge request !1

from 'key previous identifier fullname'
to 'key;previous identifier;fullname'

* Also the command line options should now work as expected
* Added an identity converter that just outputs .json output.
  This allows to postprocess files that have already been created.
* Breaking changes to the data format: students is now a list
* Bump python version to 3.5 to support typing