improve datacite field mapping and import

Current version succeeded to import a random sample of 100000 records
(0.5%) from datacite.

The --debug (write JSON to stdout) and --insert-log-file (log batch
before committing to db) flags are temporary added to help debugging.

Add few unit tests.

Some edge cases:

a) Existing keys without value requires a slightly awkward:

```
titles = attributes.get('titles', []) or []
```

b) There can be 0, 1, or more (first one wins) titles.

c) Date handling is probably not ideal. Datacite has a potentiall fine
grained list of dates.

The test case (tests/files/datacite_sample.jsonl) refers to
https://ssl.fao.org/glis/doi/10.18730/8DYM9, which has date (main
descriptor) 1986. The datacite record contains: 2017 (publicationYear,
probably the year of record creation with reference system), 1978-06-03
(collected, e.g. experimental sample), 1986 ("Accepted"). The online
version of the resource knows even one more date (2019-06-05 10:14:43 by
WIEWS update).

1 files changed, 0 insertions, 0 deletions

diff --git a/python/tests/files/datacite_1k_records.jsonl.gz b/python/tests/files/datacite_1k_records.jsonl.gz
new file mode 100644
index 00000000..28ea6e37
--- /dev/null
+++ b/python/tests/files/datacite_1k_records.jsonl.gz