add support for Jinja in a destination node header and footer

[tomreitz]

This PR adds support for a destination node's header and footer to contain Jinja. The first row of data is passed in, so __row_data__.pop('__row_data__').keys() contains the column names, which can be used to define a dynamic header/footer. For example:

destinations:
  input_no_student_id_match:
    source: $transformations.input_no_student_id_match
    template: ./verbatim.csvt
    header: |
      {%raw%}{% for k in __row_data__.pop('__row_data__').keys() %}{{k}}{% if not loop.last %},{% endif %}{% endfor %}{%endraw%}

    extension: csv
    linearize: True

[tomreitz]

@jayckaiser I updated this so the "first row" (and the corresponding __row_data__) is only computed (to be passed into the header/footer template) if either the header or footer in fact contain Jinja (via util.contains_jinja()). So in most cases we skip this potentially expensive step.

I also found and fixed a bug in the DEFAULT_TEMPLATE which was causing a Jinja parse error "circular reference detected" when it was actually used.