Oger

- new output format: *pubanno_json.tgz*, gzipped archive of PubAnnotation JSON
- new input format: *txt_tsv*, two-column TSV <doc-ID, text>

- bug fixes in PubAnnotation output

- REST API:
* various bugfixes
* new default value `true` for the postfilter parameter
* don't allow pickle objects in user-defined configuration
* allow built-in postfilters in user-defined annotators
* background loading uses threads (concurrent.futures) instead of subprocesses (multiprocessing) -- more stable now
- new formats:
* europepmc and europepmc.zip (output)
* *bionlp* (output)
* *brat.ann*, *bionlp.ann*, *txt* (output) (individual files of bionlp and brat)
* *conll* (input and output)
- fixes to input/output formats:
* *pmc* (more robust)
* *tsv* (no newlines/tabs inside mentions)
- new normalization method: ise/ize spelling
- fixes to document/collection iteration
- evaluation: misc improvements

- new input format: *bioc_json*
- built-in postfilters for removing subsumed/overlapping matches and frequent false positives
- parameter `mesh-as-entities` for recognising MeSH descriptor names as entity annotations (*pubmed* and *pxml* format)
- UMLS CUIs (given in the termlist) included in _bioc*_, *(text_)tsv* and *xml* output formats

- bug fixes in input/output formats

- new input and output formats *pubtator* and *pubtator_fbk*
- new normalization method *mask*
- new parameters `byte-offsets-in` and `byte-offsets-out` for interpreting and producing BioC documents with byte-based offsets
- improved PipelineServer class for library usage
- fixed a security hole involving eval()
- allow specifying a pretrained sentence splitter