Refactor and/or create new benchmark comparisons using the newer kernels up to l=10

Following #9, the first version of the kernels is not as dramatic in performance difference to `e3nn` + `torch.compile`. It would be good to see if the second versions, i.e. those that go up to `l=10` might see performance benefits.

Start by reading issue #9 and locating the existing kernel benchmark comparisons. Evaluate the newer kernel versions that support l=10 against e3nn with torch.compile; done means the benchmark comparisons are updated and clearly report whether the newer kernels improve performance.

Written by the indexing model from the issue text.