Add test variants with compressible data

[crusaderky]

Closes #696
Run the following tests twice, once with uncompressible data and another with highly compressible data, to display differences on the network stack in one use case rather than another:

• test_anom_mean
• test_climatic_mean (currently skipped)
• test_vorticity
• test_double_diff
• test_dot_product
• test_map_overlap_sample

This PR increases the overall runtime from 48min to 50min.

I've deliberately not touched test_basic_sum, which is always compressible, and test_rechunk_*, which are always uncompressible, because they have already a fair amount of permutations and I didn't feel that doubling everything (with the additional challenges in readability more than in runtime) would yield a benefit worth it.

[crusaderky]

I ran an A/B tests on distributed#7593 and I'm observing a very modest (5%), but consistent speedup in test_filter_then_average. The test uses data that is compressible at 37%. The other tests do not show any kind of change - including those running on data that is compressible at 99%. The reason is that the data compressible at 37% takes 140ms per chunk to compress, whereas identically sized data full of ones takes 14ms per chunk.

I need to scrap the current algorithm and synthetically create something similar to the zarr dataset.

[crusaderky]

Holy guacamole 😱

[crusaderky]

This is ready for review and merge.
Discussion on the findings on dask/distributed#7655

[milesgranger]

LGTM. 👍

[crusaderky]

ty @milesgranger for review