That way, we read the data files only once, but produce all the outputs
we need. Together with a powerful makefile, we have a multi-invocation
with proper depedency tracking.
Everything will be regenerated though, even though just a single input
template file changed.
The alternative would be to have one dependency and invocation per
input dependency, but that will read the entire json each time.
Let's see what's faster/more useful during development.
After minor modifications to pyratemp, it certainly does the job.
What it **does NOT** do:
* multiple outputs per template/command invocation
* NICE embedding of code (like GSL can)
It will do the job nonetheless, but mako might be worth a look
