Porting processing model algorithm to C++, pt 1

[nyalldawson]

This PR takes the first step in porting the processing model algorithm across to c++.

The python design of model algorithm is fine, but suffers from confusing naming. So this is pretty much a direct port of the same design over to c++. The changes include

• all component algorithms making up the model are referred to as "child" algorithms in the API
• instead of separate classes used to handle ValueFromInput/ValueFromOutput/etc, there's a single class QgsProcessingModelAlgorithm::ChildParameterSource which handles all possible source types for input parameters to child algorithms. The actual behaviour is specified via an enum (StaticValue /ModelParameter /ChildOutput ). This will be easy to extend in future with other possible child algorithm parameter input types, such as a "calculate expression result before running child alg" type (refs QEP 84: Processing ModelerParametersDialog widget wrapper QGIS-Enhancement-Proposals#84)
• model components (e.g. the model's input parameters, child algorithms, and model outputs) now use a common base class (QgsProcessingModelAlgorithm::Component) which handles the graphical modeler behaviour of the component.

It's not fully functional yet - designing models should work without issue, but there's no support for loading/saving models, or running them....

At this stage I'm after feedback on the QgsProcessingModelAlgorithm API