Adds a CameraDirty entry to QDemonRenderNode::Flag. It already has
Orthographic, so if it should move then they can move together.

Camera now stores the previous viewport and design dimensions it
calculated projection with so we can avoid the calculation completely
if nothing changed.

After having initially enabled it.

Also some drive-by changes:

- Remove always-true nullptr check
- Only go through the loadRenderImage-branch if we were dirty

Just a small utility to save lines when doing conditional assignment
and work based on equality. Used in follow-up patches.

Includes new sourceItem property on Texture

Drive-by change

The C++ parts don't work for some reason, so right now its just QML
parts.

It's obsolete anyway. From an early attempt at file import.

import QtDemon -> import QtQuick3D
import QtDemonMaterialLibrary -> QtQuick3D.MaterialLibrary
import QtDemonHelpers -> QtQuick3D.Helpers

DemonCamera -> Camera
DemonCustomMaterial -> CustomMaterial
DemonDefaultMaterial -> DefaultMaterial
DemonEffect -> Effect
DemonImage -> Texture
DemonLight -> Light
DemonModel -> Model
DemonNode -> Node
DemonView3D -> View3D
DemonSceneEnvironment -> SceneEnvironment

Examples and Converters are ported to generate new QML API Code.

Designer additions also changed.

Enable running of autotests in master

See merge request annichol/qtdemonrender!2

Unless you explicitly state so in the DemonImage format.

Don't return [-1, -1, -1] when a position cannot be mapped to/from the
visible viewport. Even if the position falls outside the viewport, they
are still correct, and can be used in calculations elsewhere.

Only when the caller needs to hold onto a type for an extended period of
time should a copy be made.

Note: Due to an over-optimistic approach in the beginning there are some
changes here that don't give us anything, but they are left as-is as it's
too much effort to revert all. The initial approach was to fragile, so I
opted for a more careful approach to avoid having hard to find issues
in the future. There are also places where we could be a lot smarter,
e.g., if we access a ref'd member of a class often without holding onto
it, consider exposing it as a public member or add an operator to access
it without causing unnecessary ref counting.

It turns out that the first version didn't work with orthographic
projections. Rather than calculating the direction from the camera
to a position in the frustum, we instead pick two points in the
frustum and calculate the direction between those two instead.

We need special handling for scoped lights because the items they
reference for scope may not have been processed before the light.  So
light notes have to have two passes.

We need to use m_cameraNode->position rather than position(), since it
matches the transformation in globalTransform. If we don't, the distance from the
camera to worldPos will end up wrong if the functions are called after a call
to setPosition, but before globalTransform is updated (updateSpatialNode()).

...so that you can increase/decrease speed in all directions without
the need to tweak each one individually