1. The semantics of data binding is much better behaved. Constraint solving doesn't have an obvious sweet spot, where you can say that you can solve a certain obvious class of constraints. Which is why you had such a cottage industry of constraint solvers.

2. Data binding is powerful enough for most things you want to express, and for those things that it's not powerful enough for, a generic constraint solver isn't powerful enough for either (e.g. ui layout, you quickly start to want a linear programming solver or something else that's domain specific).