Fix camera computation

[hozuki]

Coordinate computation in camera motion is still buggy. May be because of LHS to RHS transformation.

[hozuki]

There are some strage values in camera curves.

For example, look at frame ~5602 of alstar. Comparing to the reference visual, the converted camera is at a different position, with completely different Y (i.e. below the stage, XOZ plane).

Coverted:

pos = (-11.06979, 3.231802, -34.16661)
rot = (-9.7438932, 164.6800598, 0.7779374)

To reproduce the reference visual:

pos = (-11.06979, ~8.0, -34.16661)
rot = (~-2.74, 164.6800598, 0.7779374)

[hozuki]

At least starting from Glow Map, scenario introduced procedural camera control (type 58).

[hozuki]

General camera orientation was fixed in commits a17b8b7 and 8a0a2de.
It requires more samples to investigate procedural camera control.