Paul, to answer your questions,
2. It is the slope of your (linear) price function. Of course it makes a bit more sense as a raw utility than transformed as ZCD, but either way you get the idea that it's large and negative.
3. Are the interactions even significant? If not this question is moot.
If they are, then models with different attribute coding and with or with out constraints are different models and they can make for differences in other attributes. In that case (i.e. the case where the interactions are significant) I don't think you're doing anything wrong, but you're going to want to choose and interpret your model carefully.
4. There's not a simple formula but there is a simple procedure. Make a simulation with two versions of your product, one with the attribute levels set to produce the relevant interaction and one without. Now vary the prices of the two products until shares are equal. This is the share equalization estimate of the value of the attributes+interaction. If you want to see what ONLY the interaction is doing for you then you'll want to simulate WITH and WITHOUT the interaction (but somehow with the attributes set in a way that should produce the interaction). There are a few ways to do this, but all of them involve using a model (the one without the significant interaction) that you know is wrong.