All mathematical statements in constructive logic have an experimental test (actually since nonconstructive logic can be embedded in constructive logic via a double negation translation, they do too). Of course this isn't always the most interesting test, since usually you are modeling some higher level concept inside the logic, and you would have a more high level test. For example for the calculation of pi you would measure the ratio between the diameter and perimeter of a circle.
Pi isn't defined by how we measure it. It's defined mathematically. Predicting how a measurement of the ratio between the perimeter and the diameter of a circle comes out is just one of the predictions we can make having calculated pi inside mathematics.
But if at some point further decimals cannot be experimentally proven to be right, at best you can say "pi is such and such for such precision" and any definition that gives the same number up to that precision must be accordingly accepted. Otherwise you need to demonstrate an experiment that uses further precision.
But we don't do that, and that's why mathematics don't have to do with experience, they are an entirely different tool that also happens to be useful in experimental science.
If mathematics doesn't have to do with experience, how do you explain the fact that when we measure pi it's 3.14?
Pi is just pi defined in a mathematical way. It's true that there are other numbers or objects which can make a prediction about the measured ratio of the perimeter to the diameter, but those are not pi. Whether those other objects may be equally valid as pi for making that prediction depends on the details. There may be reasons to prefer one to the other even if they make the same predictions up to measurement precision. We usually prefer the simpler explanation for example. This is equally true in physics and other subjects. Note that as a device for predicting the ratio of the perimeter to the diameter, pi is not perfect. Our space is curved, so for large circles the ratio will deviate from pi, and you have to use a more complicated method based on Riemannian manifolds.
"why mathematics don't have to do with experience" is too strong. Nothing you just said implies that math isn't intimately connected to empirical predictions in some way. What it shows is that contrary to Jules' earlier statements, there's no statement by statement correspondence--for any given mathematical statement, you can't find an interesting empirical prediction you associate with it.