Broke the startup of the app into a series of extension methods (and classes to contain them), rather than one long method. This means the stack tracers are easier to follow when the startup fails, and it's clearer what's happening when throughout the pipeline.
I've included a few optimisations, such as starting to rely more on DI for injecting things (like the logger) rather than the static logger, which in turn can help with testability and evaluation of the callstack.
