Benchmarks: Add all from Dask tutorial

[flying-sheep]

Different data:

1. subsets
2. sparse / dense chunks

Also all from disk.

[ilan-gold]

Actually deploying this is probably blocked by correcting CPU affinities on the benchmarking machine, but writing the code for this should be manageable otherwise.

[flying-sheep]

We now have the big dataset tests, so this might be easy-ish to do

[ilan-gold]

Goal:

Add dask use-cases to the scanpy benchmarks so we can understand performance changes.

Nice links:

1. Example benchmark: https://github.com/scverse/scanpy/blob/main/benchmarks/benchmarks/preprocessing_counts.py
2. Project we use for benchmarking: https://asv.readthedocs.io/projects/asv-runner/en/latest/index.html
3. Dask local cluster: https://distributed.dask.org/en/stable/api.html#cluster
4. Using scanpy and dask: https://scanpy.readthedocs.io/en/stable/tutorials/experimental/dask.html. NOTE: this read_elem_as_dask function in the notebook is with anndata 0.11 i.e., pip install --pre anndata

[ilan-gold]

@mikelkou See #2578 for functions that have dask vs. not