[Script] Support thread group in Tilus Script

[yaoyaoding]

In Tilus Script, you can partition the threads in a thread block into smaller thread groups and define instructions that execute only within specific thread groups. This provides fine-grained control over thread execution and enables efficient parallel programming patterns.

Example:

class MyScript(tilus.Script):
    def __init__(self):
        super().__init__()

    def __call__(self, ...):
        # Specify 4 warps = 128 threads total
        self.attrs.warps = 4

        # First level: split into 4 groups of 32 threads each
        with self.thread_group(0, num_groups=4):
            # Only first group (threads 0-31) enters here

            # Second level: further split into 2 sub-groups of 16 threads each
            with self.thread_group(0, num_groups=2):
                # Only threads 0-15 execute this
                fine_grained_work()

            with self.thread_group(1, num_groups=2):
                # Only threads 16-31 execute this
                different_fine_grained_work()

            # Back to first level - all threads 0-31 execute this
            self.sync()