Add support for scale to OrthograpicProjection

[DJMcNab]

Example:

Before this change, it was impossible to get a larger (or smaller) image from an OrthographicProjection - this is what the new example looks like with scale set to 1 (old behaviour):

Notes:
The scale could be inverted, but this interpretation also makes sense
Maybe there's a better name than scale - what do other engines call this?
Maybe it could use the Scale component - would require passing a scale parameter to CameraProjection::update, which would be meaningless for PerspectiveProjection, unless it effected fov somehow

[MichaelHills]

If you used a Vec3, aka NonUniformScale (or supported both?), would that let you flip Y? I think right now, X is right and Y is up but I wanted to flip it so Y is down.

[DJMcNab]

The issue with that is how would NonUniformScale work for PerspectiveProjection?

[MichaelHills]

I was purely talking about OrthographicProjection as that's where you added the scalar scale.

[chemicstry]

I believe you can always flip, scale, rotate by just camera Transform?

This PR makes it easier by detaching projection from screen size. You could do the same with existing orthographic projection, but you would have to constantly rescale camera Transform based on window size.

[chemicstry]

Sorry I mixed this with other similar PR #400. I think it is better, because you can adjust scale with just Transform, no need to introduce additional variable.

[MichaelHills]

Thanks I was able to use Transform to flip Y (and ran into face culling issues because the winding order flipped!). Just using Scale here makes sense.

[chemicstry]

What Transform did you use? Maybe you inverted y?

I just tested horizontal and vertical flipping by setting x or y scaling to -1.0 and there were no issues. Rotation around y should work too

[DJMcNab]

I've experimented some more with this, and discovered that using a scale does work, thanks to

bevy/crates/bevy_render/src/render_graph/nodes/camera_node.rs

Lines 120 to 121 in a5f6cb0

	let camera_matrix: [f32; 16] =
	(camera.projection_matrix * transform.value.inverse()).to_cols_array();

However, the actual problem is with the clipping planes. I do think the whole projection scheme could do with some work to fix the clipping issue (see also #400).