Refactor to use BigQuery syntax

[alistairewj]

As discussed in #717, it will be easier to maintain one set of scripts rather than two branches.

I've updated all the scripts to use BigQuery syntax, and rewritten everything to be compliant to both BigQuery and PostgreSQL. BigQuery tends to do date arithmetic using purpose built functions (DATETIME_ADD, DATETIME_SUB, etc). I wrote PostgreSQL functions which emulate this functionality. Unfortunately, it's not possible to emulate DATETIME_TRUNC or DATETIME_DIFF, as these have a BigQuery primitive date unit as an argument (e.g. DATETIME_DIFF(date1, date2, HOUR)), and it's not possible to have PostgreSQL recognize anything but numeric values as a primitive in function arguments. I've worked around this by having a shell script (postgres_make_concepts.sh) which calls the functions, and it uses a regex to single quote the dateunit (i.e. HOUR -> 'HOUR'). It also changes the table names from physionet-data.mimiciii_clinical.<table> to <table>. Works well enough - though unfortunately will cause a headache for Windows psql users.

The make-concepts.sh script can be used to rebuild all the concepts on the mimiciii_derived dataset on BigQuery. The script will generate tables on whatever the default project is - for us this is physionet-data.