Revisit handling of blending around shadow blobs in SfS

[oleg-alexandrov]

In SfS, a region that is in shadow is considered to have no useful data, so SfS does does not happen in it. Then there is a transition area around it to terrain where SfS happens, to avoid a discontinuity. This results in some erosion or loss of detail around the region.

When many such small areas are present, the transition areas can end up eroding a lot of the good terrain produced with SfS. It is then preferable to do SfS in such a small hole and live with a little artifact than skip and erode around it.

This is controlled by parameter --min-blend-size. Regions in shadow whose both width and height are less than this are considered acceptable to do SfS in. For bigger shadow regions, SfS is not done.

The question is, is defining a region to be small by its width and height the right thing? What if there is a long and narrow shadow? Maybe it is better to do SfS there too, and not erode around it.

This will be looked into when there is a solid testcase to illustrate the problem.

[oleg-alexandrov]

This could also like be tackled by @AndrewAnnex, if @rbeyer (ticket originator) can find justification under current funding umbrella.

The work will amount to adding an optional argument to a function and passing it a few layers down. I can share details.

Testing could be done on a small but representative subset of some recent production SfS work (say 4 km x 4 km in an area with many shadows).