Significant increase in number of tasks of test_climatology::test_highlevel_api

[fjetter]

I just noticed that dask/dask#11479 multiplied the number of tasks by 2.5x before and by 4x after optimization. I haven't looked at what this does to the end to end runtime but initial submission and client/scheduler side memory obviously takes a hit. Given that before/after optimization task number ratio got much worse I suspect we're loosing opportunities for trivial fusion due to this rechunking

# before https://github.com/dask/dask/pull/11479
Enter optimize 210477
After optimize_blockwise 166163
After fuse_roots 166163
After cull 77002
Low level 77002
After fuse 71878
After inline 71878
After opt slices 71878

# after https://github.com/dask/dask/pull/11479
Enter optimize 535113
After optimize_blockwise 390839
After fuse_roots 390839
After cull 301678
Low level 301678
After fuse 293482
After inline 293482
After opt slices 293482

[hendrikmakait]

I think this is expected. We should probably compare the current task count to something like the workload before #1583.