CNT contextual layers

#20 is about adding contextual layers to help the user both initially pick their neighbourhood boundary and do designing inside. That issue will be about implementing this generally for anywhere in the world, using OSM and other global datasets.

In the short term though, we're going to start this work focused just in Scotland. The initial list mostly comes from two existing projects that should be quick to start adapting here.

From https://nptscot.github.io/npw/ / https://nptscot.github.io/:

- [x] GPs and hospitals
- [x] Schools
- [ ] Greenspaces
- [x] Existing cycle infrastructure (blindly trusting the NPT pmtiles file)
- [x] NPT flows (with full filtering / styling controls)
- [x] Estimated traffic volumes
- [ ] Gradient
- [x] Current level of service (blindly trusting NPT data, not calculating it)
- [x] SIMD aka deprived population
- [x] Population density

From https://acteng.github.io/atip/browse.html:

- [x] Railway stations
- [x] Bus routes
- [ ] Bus stops, with frequency of use
- [x] Stats19 collisions
- [x] Car ownership

Small followup tasks:

- [ ] Shrink size of GP, hospital, and school GJ files

The issue references existing projects (NPT and ATIP) as sources for contextual layers. Start by examining how these projects serve their data (e.g., pmtiles, GJ files). Look at the codebase for existing map layer integration and data processing. The task involves adapting and adding specific datasets (greenspaces, gradient, bus stops, etc.) for Scotland, then optimizing file sizes for GP, hospital, and school data.

Written by the indexing model from the issue text.