Templatize Point2/PointLL

[danpat]

Issue

The Point2 class, (and PointLL that is subclassed from it) is hard-coded to use float.

Most of the time, this is fine. But sometimes, you want to do something that requires a bit more precision. In my case, I'm splitting geometry using midgard::trim_polyline.

This leads to creating new points along existing lines. When using float, practical precision is limited to about 0.5m wrt geospatial coordinates. Most of the time this is fine.

However, in my case, I'm rendering a map with the original geometry, plus some newly injected intermediate points created by midgard::trim_polyline.

When float precision is used, sometimes the error margin becomes visible in weird ways, like in the circled section in the following screenshot:

Using double for the newly created types has enough precision that they are "on the line" for all human-practical detail levels - float does not, and half-metre errors do become visible in some real-world contexts.

Long story short - this PR turns Point2 into a template class, and makes all the necessary changes to support that. Most code remains unchanged, and float precision remains the default, but this refactor allows you to optionally choose PointXY<double> or GeoPoint<double> (for spherical Distance, etc) if you need the extra detail for some reason.

Tasklist

• Add tests
• Add #fixes with the issue number that this PR addresses
• Generally use squash merge to rebase and clean comments before merging
• Update the changelog

Requirements / Relations

Link any requirements here. Other pull requests this PR is based on?

[danpat]

@kevinkreiser @purew This is ready for a round of 👀

[purew]

LGTM 🍰