Terrain brush fixups

[BogdanW3]

• Use DP to optimize the apply call's updated area
• Fix the 3x3 kernel for cliff_near
• Unflip the brush's display

[stijnherfst]

Rewriting it to use a stack based approach is really good. I would maybe make some suggestions on the design.

Instead of taking a std::vector<glm::ivec2>& seeds the function would be more general if it purely used the QRect area as input for what area to check.

std::vector<uint8_t> visited(static_cast<size_t>(corner_width * corner_height), 0);

Is a bit pessimitic in allocating for the full map while most operations will likely never even trigger any cliff changes. A hashmap (hive::unordered_map is quite optimized too) will likely be better in most cases with only the worst case (whole map change) being a bit worse.

visit_index is named like an action but it's more like get_linear_index

Nice catch on the bugs :)

[BogdanW3]

Thanks for the reply! Indeed it sounds like I can just use the area as an input as well, especially with the suggested change to a hashmap, but sadly my attempts to do that haven't been successful thus far.
Even without that, it seems there is a bug with the clamping logic where I am sometimes somehow able to make a specific cliff glitch out, I've had more luck doing it by first adding +4 to the base terrain then doing -4 in a small section and then dragging it around.