Fixing issues with LookAt cameras

[davesmith00000]

I started this work to fix #622, as reported by @Akkete. This was a graphical glitch that showed gaps between clone tiles when using a look-at camera.

As expected, the issue was simply a rounding error which I replicated in the test scene in this PR, and then fixed.

However, while I was doing that I also noticed that LookAt camera's were behaving weirdly when you have different levels of magnification going. Essentially, layers are all rendered at 1:1 and then stretched up to the right size (this helps give nice crisp pixels and reduce's rendering requirements quadratically) and so the projection matrices were being calculated with a magnification of 1 irrespective of what the layer / scene update fragment had to say. This works fine for fixed cameras, but look-at camera's need the real magnification for their matrix calculations.

No doubt this just slipped the testing net in days of old, and I haven't used a look-at camera since. It happens. 😬

I've fixed both issues at once, and tested this in a variety of combinations (this time...), including WebGL 1. The solution is a bit... ugly. The CameraHelper should probably be refactored. It currently manages to produce the projection matrix for several possible scenario's at once, and is a bit spaghetti-like... but it works.