The old Gain Twist Barrel Company did, but they folded up. I don't recall who did it first, but the idea has been around for some decades. AFAIK, it has never proven to help bullets or accuracy. I think the idea has two problems: First, it ignores that the bullet is accelerating in the barrel anyway, so increase in rotational velocity occurs at the same rate as for the bullet's velocity in linear rifling, anyway. Second, Harold Vaughn has shown that core separation and slip is a real phenomenon when you give a bullet too much rotational acceleration. Once the jacket gets spinning faster than the core, after exiting the barrel it is slowed by friction with the core with the result you get less net stabilizing spin than the twist rate should give you. You also get a jump as the masses align on a sometimes-new axis, so accuracy is deteriorated. If you leave the majority of your rotational acceleration of the bullet for the far end of the barrel to accomplish, the bullet is already going pretty fast when it gets to the more rapid twist, so peak rotational acceleration is increased. That results in core separations at a lower muzzle velocity than would occur in a conventional barrel.
A non-linear twist function could be devised to make rotational acceleration truly constant, I suppose. It would be different for every powder, charge, and bullet, however, which makes it a difficult thing to apply in practice.