Prioritize by density

[michaelkirk]

FIXES #154

This is based on #187, so review that first.

We still need to adjust the buckets (#143) - it looks pretty washed out, even in Glasgow, but over all, I feel like this fits into our existing patterns pretty well!

You'll no doubt noice that I'm using our existing population zone data - the ones we're using for SIMD. I understand there may be finer grained population geometries we could use.

I think it may be worth investigating using these smaller geometries, but consider:

1. It's additional geometry data we'd have to send the user, and store in their RAM
2. It's more computation — we'll have to Prepare and Intersect an additional set of population Geometries rather than piggy-backing on the Intersection calculation we're already doing for SIMD.
3. Ultimately, it might not materially affect the output all that much, since the existing population zone geometries are already reasonably to scale with our LTN boundaries.

My suggestion is to make a follow-up issue, and prioritize it with respect to everything else we want to get done, but what do you think?

[dabreegster]

Starting with the data zone level data is the right call. If we get feedback that output area is necessary and we have time, we can try it and see what the perf impact is, but this looks totally fine to me.