Oger

- new switch "iter_mode" for explicitly choosing iteration mode (PMID vs. directory vs. collection), rather than a flag for overriding an implicit default
- new iter_mode "collection", which produces a single output file for a collection of articles
- faster term look-up (using two dicts now)
- term matching is based on a special (non-NLTK), customisable tokenisation
- additional termlist fields are exported to TSV and XML as well
- separate flags for controlling output details (include\_headers in TSV, sentence_level annotation in BioC), instead of the overloaded pretty\_print flag
- encoding: rather than being controlled by the locale, encoding is now always UTF-8 for all non-XML input and output documents

- flexible term list: extra columns can be specified through an option (`termlist_extra_fields`)
- articles can be processed in directory mode, ie. the input set is defined through directory contents (with optional glob filtering) rather than through a list of PMIDs.
- faster PubMed download by requesting multiple articles at once (up to 1000 by default); also the dependency from BioPython is removed in that the efetch API is directly accessed with urllib (however, the pickle-caching was dropped)
- multiple output formats per run can be specified
- new output format "Brat" (creates a plain-text file along with a standoff annotation file)
- new input format "pxml.gz", which parses the Medline chunks (gzipped PubMed abstracts)

- new input format: BioC (reads and writes at the collection level)
- PMIDs can be specified in the INI config file, even when running the pipeline from run.py.
- default-settings.ini is read by default, if present (that way, the pipeline can be configured to be run without command-line args, while still avoiding conflicts with upstream repos)
- run.py is easier to use as a module through the top-level function `run()`, which accepts any keyword args and runs the same way as in script mode