Refactor CLI code to separate user input from execution

[ryardley]

We should tidy up and refactor the CLI code to separate layers:

• user input.
• validation.
• execution.

Why should we do this?

• Being able to mock the input layer will mean we can easily test the cli validation layer and assert against the results.
• We can replace the top layer with passed in values removing all interactivity.
• We can assume that if our validation is good enough that execution will likely succeed.
• Consistency / Readability