Add SQL intermediate representation

[skalee]

This pull requests adds a SQLite3 intermediate layer. Concepts are first imported from spreadsheet to the SQLite3 database, and then exported from the database to YAMLs.

Two CLI commands are introduced: xlsx2db and db2yaml. The xlsx2yaml which existed before is preserved, though now it uses SQL for intermediate data representation.

Motivation

• I need some data analysis tool, and for many reasons SQL suits me better than spreadsheets.
• Now I can convert the spreadsheet to SQL, and then use that SQL source for testing concepts processing. This saves a considerable amount of time at each iteration, because parsing spreadsheets is really slow.
• Thanks to SQLite3 console, it's easy to cut an arbitrary subset of input data to work on.
• Database constraints already helped in anomaly detection, e.g. #93.

Performance considerations

• The intermediate layer adds some overhead when converting from XLS to YAMLs, but it's insignificant (2% slower on my computer).
• SQLite3 increases memory usage, but the database size is just about 50 MB.

Final notes

• This pull request adds SQLite3 as a dependency. I can make it optional if someone is unhappy with that.
• The result of xlsx2yaml is slightly different in case of duplicated rows (see #93). It used to take the last row for given term ID and language pair, but now it takes the first one.

[skalee]

Ouch, I wanted to create a pull request. Closing in favour of #96.