Investigate potential inaccuracies in the surface normal generation for sloped terrain geometry

[rdw-software]

One of the issues I've found while investigating #345 and #352 is that the surface normals are clearly wrong (at least for slopes). It's unclear whether this is a big problem; it might explain why diffuse colors aren't rendered correctly, but it may also be a red herring. In order to investigate the impact of this, a flat terrain map should be created, by simply setting all heights to zerot:

• If this fixes the inaccuracies observed in diffuse texture blending stage (simulated in the terrain shader), normals are the issue
• If this doesn't fix the colors being "too dark" in pay_dun00 then there may be yet another problem that needs investigating
• It should also (always) fix the shadows being projected to the walls of pay_dun00 and the slopes in xmas_fild01
• Given that the rendition looks better for sloped terrain (IMHO), it's only worth "fixing" if it does cause the wrong diffuse colors

I'm fairly certain that they took some shortcuts here to generate less accurate normals faster, which explains the delta.

[rdw-software]

This doesn't meaningfully change the output of the lighting model; Flattening the terrain doesn't resolve any of the remaining rendering issues, which makes me think that this is more or less completely irrelevant (within the acceptable margin of error).

[rdw-software]

Looks like switching the output format from sRGB to RGBA fixed all remaining issues - at least as far as I can tell.

[rdw-software]

Hmm... I think this was actually unrelated to the sRGB output; there are some inaccuracies on sloped terrain still. IIRC replacing the diffuse textures with a white (blank) image and disabling all other rendering features resulted in visible differences for the interpolated colors from the directional light source. I also feel like there are visible differences on xmas_fild01 (blue-ish tint).

No big deal, but I'll reopen this nonetheless. First, should try to reproduce the original issue in a flattened pay_dun00 though.