(I turned my comment into an answer)
I don't have any published research for you but the reason we generally don't ask those things in interviews is because interviews tend to be geared towards qualitative research whereas, the questions you mentioned are more quantitative.
Usually, in an interview, you are testing a hypothesis (Like "Can the user find how to purchase items easily?"). You will be asking them questions about how they read the interface, what they are attempting, why the might be attempting it in that way, etc. You'll also be conducting only a few interviews like this.
Asking the candidate what they want is no longer testing a hypothesis - It's canvasing opinion. For that to have any kind of statistical significance you need to get responses from a large percentage of your users. You might interview five people who, by chance, all come from roughly the same socio-economic background, for example, but the majority of your users might come from a different socio-economic background. Therefore the opinions of the interview group would not necessarily reflect the opinions of the majority of you users. (It doesn't have to be about socio-economic background. There are infinite ways your interview group could align differently to the majority of your users).
You need a much larger sample when canvasing for opinions.
It's not that you shouldn't ask what users want during interviews, just that the information you get from it will be worthless.