`LaunchConfig.grid` is in the wrong unit if `.cluster` is set

[leofang]

When comparing notes between cuda.core and cccl-rt @pciolkosz and I noticed one discrepancy. The public example code we have

cuda-python/cuda_core/examples/thread_block_cluster.py

Lines 67 to 71 in 127f798

	# prepare launch config
	grid = 4
	cluster = 2
	block = 32
	config = LaunchConfig(grid=grid, cluster=cluster, block=block)

is meant to match the C++ counterpart

cudax::make_hierarchy(cudax::grid_dims(4), cudax::cluster_dims(2), cudax::block_dims(32));

However, it actually does not because cuda.core generates the following output today:

$ python thread_block_cluster.py 
grid dim: (4, 1, 1)
cluster dim: (2, 1, 1)
block dim: (32, 1, 1)
done!

which is wrong. The reason is that we did not do a proper unit conversion internally. When .cluster is set, .grid = 4 should mean we have 4 clusters, not 4 blocks. The correct output should be

grid dim: (8, 1, 1)
cluster dim: (2, 1, 1)
block dim: (32, 1, 1)