RefExtract: improves docextract, refextract

• Bumps version to 1.5.32
• Adds "0" subfield that associates recids to refs.
• Improves arxiv rn parsing.
• More permissive url, authors and doi regexps.
• Reworks pagination removal.
• Reworks journals identification.
• Reworks references splitting.
• Reworks refs section detection.
• Handles authors with weird characters.
• Fixes -c option.
• Prevents have 2 "a" (author) subfields.
• Improves handling of PoS.
• Removes "xbook" subfield.
• Adds "c" (collaboration) subfield.
• Only accept 1 as first ref.
• Only output o subfield if marker is present.
• Tweaks find numeration.
• Only create tickets for records that are less than 2 years old
• Only accept 1 as first ref.
• Only output o subfield if marker is present.
• Fixes marker detection.
• Allows : and - in line markers.
• Tweaks find numeration.
• Fixes find rawref The tag number for converting to the journal format was hardcoded This is not the case anymore.
• Handles ATL-CONF: Hande several formats for atlas conferences used in references ATL-CONF-99, ATLAS-CONF-2010-001, ...
• Fixes missing lines when a line marker with a number that. does not match the sequence (1, 2, 3, etc.) is detected.
• We do not strip pagination anymore. I am keeping this test in case we re-enable it.
• Fixes bug that would remove letter from o subfield
• arXiv report numbers with a category are not found by the search engine so we remove the category before searching.
• Script to convert journals abbreviations to short form
• New record model similar to bibrecord but easier to use
• Switches to the new bibrecord model to generate xml. This way we can generate it more easily and we can ensure it will always be valid.
• Encoding tweaking: we now return bytes instead of unicode for functions that provide a serialization format namely xml
• If a reference has 2 different DOI, we split it We also remove duplicate doi tags.
• Prevent authors with "paper" as name
• Handles 1.1 numeration
• When unidecode produced a string with a different length than the original, the replace would be off overwriting text to the right. We know pass the output through unidecode in that case.

Signed-off-by: Alessio Deiana <alessio.deiana@cern.ch>