Clamp unproject to mercator bounds in all projections

[mourner]

This PR fixes the issue where the map breaks when you pan outside the world in alternate projections — it does so by adding clamping logic to each individual projection so that unproject never produces NaN or values outside of the Web Mercator bounds (-180, -85.05.., 180, 85.05..), and you can't pan in a way that puts center of the map outside of valid geographic bounds. Notes:

• Clamping has to happen inside individual unproject (sometimes in the middle of calculations, e.g. in Albers) because you can't clamp once a value turns into NaN.
• Theoretically, some projections could be clamped beyond Web Mercator limits (e.g. 90 latitude and not 85.05+), but it might be better for consistency to clamp to the same range everywhere, especially since our vector and raster tiles only cover the Web Mercator range, and panning outside it might be counterintuitive.
• Projections that can have world copy rendering only clamp lat, not lng, to keep the current behavior of panning seamlessly through anti-meridian (which doesn't yet work for wgs84 projection but keeps the default behavior of mercator).
• There's a known issue of the map going wild jumping chaotically during fast panning inertia. Not critical to address in this PR but we should ticket this and follow up.
• This approach feels pretty intuitive. You can't move the center of the map outside the rendered area, but you can still grab the map in empty areas for panning. getCenter()/etc. will always return valid results. And we avoid the issue of breaking the return type signature when porting this to native.
• In the future, we should introduce wrapping for Albers so that it could represent areas like western islands of Alaska & far eastern Russia seamlessly instead of having a hard cut (e.g. so that "anti-meridian" moves to a different vertical tile bound).

[ryanhamley]

I'm seeing the projections jump from the left edge to right and vice versa. I understand why this is happening based on the clamping and wrapping, but it seems like undesired behavior, at least when world wrapping isn't turned on.

[mourner]

@ryanhamley nice catch! Fixed this in the commit above. Generally, I should work on overhauling our constraining / bounds / wrapping / worldCopy logic. Let's leave that for a separate PR though, so that we can merge this more minimal one that at least gets us to a state where map doesn't break on interaction.

[ryanhamley]

This looks good, but there's one other thing I'm seeing. Go to http://localhost:9966/debug/projections.html#2.6/25.98/-83.31/35.7 for example (in the default Albers projection) and you can see that it's impossible to pan the map so that the southern tip of South America comes into view. The same happens at http://localhost:9966/debug/projections.html#2.6/4.02/154.71/179.1 where you can't view the upper left corner of the map. I know realistically, there's no reason to center an Albers projection on the US then try to view Patagonia, but it feels weird to not be able to pan to certain parts of the map. Do we just accept this behavior and document it (and maybe note that conical projections should generally be used with a map bounds)?

[mourner]

@ryanhamley No, that's what I was going to look at next in a separate PR as mentioned above. It's the constraining logic in transform which currently doesn't take projections into account.

[ryanhamley]

Sweet. This looks great then!

[ansis]

Web Mercator bounds (-180, -85.05.., 180, 85.05..),

This is going to change existing behavior for map.unproject(...) in mercator, right? I think this currently returns values outside this range for both latitude and longitude. I'd say the clamping should be to whatever makes sense for each projection, which I think means no clamping for mercator.

I think all of the projections implemented so far should be able to support the full latitude range