Reduce page weight by ~65MB

[slowe]

I noticed on the intro page a note saying "there may be a slight delay in loading the interactive map depending on the device you are using". Looking at the interactive map page I noticed that it has a 97MB page load (uncompressed) of which 89MB is in the HTML. Looking at that I noticed most of this is in inefficient JSON. In this pull request I've:

1. Truncated coordinates from 15 decimal places (which is equivalent to atom-level precision) to 5 decimal places (around 1.1m at worst which should be fine for 200x200m boxes). That saves approximately ~21MB;
2. Converted RGB values into integers instead of floating point values. This saves another ~1.7MB.

These were a couple of "quick win" improvements. However there are is potential for a lot more reduction:

1. There seem to be 195,395 cells so the variables accessible_population, id, population, proximity_population, transport_performance, uc name and within_urban_centre are repeated that many times. Are they used in the visualisation? If not, they could be removed entirely saving another 27MB. If they are, could they be shorter?
2. The quadrilaterals that make each cell are currently sent as GeoJSON in the JSON. That requires 10 values to be sent for each cell (lat/lon for 5 vertices). With a little re-engineering you could send a more concise format to the browser. You could then use JS to construct the GeoJSON representation before sending it to the map. I suspect that 195,000 operations may be much quicker than the many MB of HTML required.

[slowe]

I've done the two extra points:

1. saved the polygons as 4 vertices (rather than 5) in geom and rebuilt them as geometry objects in the browser - saves 11MB
2. trimmed variable names - saves 15MB

[slowe]

I've converted the data variable for each layer from JSON to a CSV format. This saves another 15MB as we don't need to repeat variable names. There is a little bit of Javascript to parse this back into the original format on page load.