Prepare for Airflow deployment

Highlights:

• fuse scored topics with null ones
• test final output
• add pipeline docstring with a link to the expected output schema
• no explicit output partitioning. It worsens performance and isn't needed, since downstream jobs are expected to read all the data all the time
• make I/O HDFS paths relative to a working directory
• optionally pass the working directory to the CLI, defaulting to section_topics relative to the current user home
• update CLI