If you are seeking inter-coder consistency – that is, multiple people coding the same sets of interviews whose coding is more-or-less consistent among them – you would want to develop a basic code book beforehand, or at least after the first one or two interviews done with open coding.
If you are taking more of a naturalistic inquiry approach, you would code your interviews using open coding – that is, developing the codes as you read through the interview transcripts – and build your code book as you go.
I’ve not heard of coding the initial questions (since you’ll be asking a whole bunch more than your prepared questions during a semi-structured interview). You may choose to code the questions you actually ask in the transcripts if you believe they are salient to revealing information from your participants. For example, if you probe a topic one way and get one type of answer, and probe it another way and get a different type of answer, and you feel that this distinction is important in your analysis, you may want to code the questions. But that would more likely be done post-hoc.