Manage the tree-build pipeline with DVC

[jaredkhan]

https://dvc.org/

I wanted to be able to script the construction of the main tree so that it could all be run in one command.
Of course, many of these steps are quite time consuming so you don't want to have to re-run them if they've already been run recently. Enter DVC ('data version control'). It's a little tool for letting you cache files based on their hash (like Git LFS) and also setting up pipelines for reproducing all these files. So far, I've set up a pipeline for producing the mainline AllLife outputs.

Source configuration files (such as the bespoke tree files) remain tracked in git.
Derived files (such as the filtered wikipedia titles files) become tracked in DVC and their reproduction is now as simple as:

dvc repro --pull

All the download stages have been automated to make this work.

The underlying logic being run has for the most part been left untouched, though some paths have been shuffled around to make the caching tidier.

I wanted to put it up for review at this stage and see if we're keen on moving forward with this.
I've set up a bucket for mucking about with this and I'll share the keys on the Slack channel. If we want to go ahead, we should set up a bucket that's under the orgs control rather than mine.

[jaredkhan]

Marking as draft only because the bucket should be changed and we should get @davidebbo's eyes on it before it's ready

[davidebbo]

What about the step where it gives you SQL commands to run to update the DB (steps 6 and 9 in the manual steps)? Isn't that still needed with DVC?

[jaredkhan]

Superseded by #123