When camera.roll = 0, the camera.changed event doesn't fire when changes only occur to the heading. e.g. if the user is looking down and rotates the camera. This causes issues with our compass not updating.

Sandcastle example:

Try the rotate buttons in the sandcastle below, you'll see that the camera.changed event is fired only when the roll =/= 0:
[Sandcastle](https://sandcastle.cesium.com

Please respect the following template to report a new issue, otherwise your issue will be automatically closed.

Before submitting, search through existing open or closed issues reports and through blender.stackexchange.com to check whether the issue already exists.

If the issue concerns an operator that grab geodata from an external web service (Basemaps, OSM, SRTM), then it can be a network

maptalks's version and what browser you use?
v0.43.0, chrome

Issue description
Layers do not share WebGL context, which creates glitches when a map has more than one 3d layer. For example a buildings layer with a 3D line will not hide the line behind the buildings.
In particular, I am implementing a 3D terrain layer but this layer will not hide (behind mountains) any other 3d element

Is there a way to know when the imagery was collected? I assume sometime in the daylight hours, and probably recently(?) but I have no way of knowing by looking at the output and the associated image.

Can that output be put in the CLI output or some other metadata file that gets associated with the LC####### directory?

rasterio.warp.reproject takes a kwargs parameter, which is passed to the "transformation function". It is not by any means clear what this means. What transformation function? What are the allowed values for kwargs? Can this please be expanded on? Thanks.

What's the problem this feature will solve?

I am currently working on a project using mesa. Great framework! The problem is that I use ContinuousSpace, and want to visualise it. Is this something you are working on?

When I edit a polygon with a zoom level low and I snap this polygon to another one, it's very common the created snap point is not exactly on the line / point.
For me, it's a problem regarding I control if these polygons are not intersecting.

An option like $ fio cat --bounds w s e n --bounds-crs <crs> --dst-crs <crs> would allow for filtering prior to reprojection, like $ ogr2ogr -spat w s e n -spat_srs <crs>.

See Toblerity/Fiona#210 for a discussion about standardizing bounding boxes as --bounds w s e n rather than --bbox w,s,e,n.