[IR][Script] Allow specify cluster dimensions through cluster_blocks

[yaoyaoding]

This PR allows Tilus Script to specify the cluster dimensions, a feature occured in Hopper and later GPUs.

class Kernel(tilus.Script):
    def __call__(self):
        self.attrs.blocks = ...
        self.attrs.cluster_blocks = [2, 2, 1]  # must be compilation-time constants
        self.attrs.warps = 4