Surface plots do not quite follow the data

[alexcjohnson]

See https://codepen.io/alexcjohnson/pen/gKLBgR?editors=0010

I was trying to get the contour lines to exactly match the data, but notice that the line at y=1.3 (last black line in the red) is not quite straight (even though the data at y=1.3 is uniformly 1.0) and the line at y=1.4 (first black line in the green) does not quite reach the correct value of 0.898 like its neighbors at larger y do. Digging in a bit deeper, it seems like the contours are following the surface very closely - perhaps not precisely, there are still a few spots where they seem to disappear behind the surface, but it's pretty close. But the surface itself seems to disagree a little bit with the data, which you can see by zooming in near one of these corners:

It looks like either a) the surface is smoothing a little bit around these sharp corners, or b) the surface is actually creating a finer mesh than the data points, and this mesh is not aligned with the data so it misses the corners. Either way it's not doing the right thing.

cc @etpinard as mentioned in #2712

[archmoj]

Here is another demo to illustrate this issue.

[archmoj]

One of the best illustrations for this bug is gl3d_surface-lighting mock where those high points on the diagonal have z values equal to 1!

[archmoj]

Here is a simpler demo for debugging this issue.

[archmoj]

When this is fixed gl3d_surface-lighting should look something like this:

[alexcjohnson]

An extreme example of the problem - spiky data like gl3d_surface-lighting (the data is only 0s and 1s) but very nonuniform x/y data - https://codepen.io/alexcjohnson/pen/XyBReJ?editors=1000

notice that the thinner spikes rarely reach 1.