special_cases::harvester::ignore_processed_files should allow for in-list reference merge

[shippy]

In special_cases::harvester::ignore_processed_files::[site] (introduced in #341), we use a YAML reference to an array of files. At the time, I thought this would be both DRY and easily extensible - YAML can use references to merge dicts, so why not arrays?

It turns out that YAML can't merge arrays, at least not in a way that results in a predictably flat results. (- *DUPLICATED_PASAT_FILES translates to a list of lists.) Under the current structure, you can only extend the referenced array, or you're out of luck.

A workaround is a value-less dict that looks like this:

harvester:
  ignore_processed_paths:
    _duplicates: &DUPLICATED_PASAT_FILES
      ? E-01099-M-7-2017-09-19.csv
      ? E-01318-M-2-2017-08-28.csv
    ohsu:
      pasat: 
        <<: *DUPLICATED_PASAT_FILES
        ? OHSU-specific-file.csv
    sri:
      pasat: 
        <<: *DUPLICATED_PASAT_FILES
        ? sri-specific-file.csv

After parsing, this comes out as

{'harvester': 
  {'ignore_processed_paths': 
    {'_duplicates': {'E-01099-M-7-2017-09-19.csv': None,
                     'E-01318-M-2-2017-08-28.csv': None},
      'ohsu': 
        {'pasat': {'E-01099-M-7-2017-09-19.csv': None,
                   'E-01318-M-2-2017-08-28.csv': None,
                   'OHSU-specific-file.csv': None}},
      'sri': 
        {'pasat': {'E-01099-M-7-2017-09-19.csv': None,
                   'E-01318-M-2-2017-08-28.csv': None,
                   'sri-specific-file.csv': None}}}}}

This should only require a straightforward change to config_utils.py::flatten_path_dict, which checks if val is None, and if so, output.append(new_prefix).