* masked walls were inserted into the sprite render list, causing sorting issues and depth fighting (see Duke's cameras and the 'closed' sign in Route66/Carnival.map)
* vertical and horizontal wall sprites needed their sorting functions swapped.
* plane clipping for wall sprites used bad coordinates.
This was by far the messiest game, there's two reasons for this.
First, the portal links do not need to be in an actual portal sector, so they cannot be used to detect portal sectors.
Second, the game moves portals in place, so all offsets are (0,0,0) so that not even these can be used for detection.
The only working method is the super-complicated original way to look up portals at run time, just being done at map start.
Having static portal links should reduce the render glitches quite significantly because the renderer knows now which sectors belong to a portal and can use this information to ensure proper processing.
Unlike the other games these are so poorly defined that the engine has to rely on the original fudging to pick the proper portal to link to. As a result they are just as limited as they always were.
In addition all the portal search code had to be reinstated.
Note that these portals cannot be rotated, so to preserve bad setups this limitation needs to be replicated.
If we want proper line to line portals for all games they need to be made a different portal type.
* do not check the clipper in the collection pass to reduce number of bunches. Clipping here brings no performance gain.
* fixed the loop reset in FindClosestBunch to actually process the array's first element.
Not tested yet and somewhat stripped down, as the portals this needs to support are a lot less complex - plus some of the issues in Doom do not apply here.
The checks in CheckTexturePosition were not correct and caused unwanted texture shifting.
Also inlined GetYPanning into DoTexture because the pow2size value is needed there and the rest is a simple one-liner to scale the panning.
y-panning is still not correct for all cases but better than before. The aspect factor is not needed, it looks like an artifact of both the software renderer and Polymost.
We get the same dimensions with proper matrices by setting this to 1.
