fix: use dask to speed up cxg conversion

[Bento007]

Reason for Change

• During the schema 5.2 migration it was found the cxg conversion was taking exponentially longer for larger datasets. This work makes use of the Dask library and anndata 0.11.0 to speed up conversion.
• This PR is not intended to provide memory optimizations, but some improvements have been observed in testing.

Changes

• add dask dependecy for propessing
• update anndata dependecy for processing
• modify cxg conversion code to use dask to write tiledb arrays.

Testing steps

• TODO

Checklist 🛎️

• TODO

Notes for Reviewer

• Do not merge until the offical release of anndata 0.11.0

[github-actions]

Deployment Summary

• backend_url: https://pr-7364-backend.rdev.single-cell.czi.technology
• backend_de_url: https://pr-7364-backend-de.rdev.single-cell.czi.technology
• backend_wmg_url: https://pr-7364-backend-wmg.rdev.single-cell.czi.technology
• frontend_url: https://pr-7364-frontend.rdev.single-cell.czi.technology
• delete_db_task_definition_arn: arn:aws:ecs:us-west-2:***:task-definition/dp-rdev-pr-7364-deletion:1
• migrate_db_task_definition_arn: arn:aws:ecs:us-west-2:***:task-definition/dp-rdev-pr-7364-migration:1

[Bento007]

Memory test result

dataset	main	dask
10x
slide seq
visium

Overall the speed improved across the board. The remaining memory spike is cause by tiledb.consolidate

Peak Memory

dataset	main	dask
10x	9021.359 MiB	8481.504 MiB
seq	8847.473 MiB	8811.594 MiB
vis	8642.473 MiB	9412.105 MiB

The memory usage was stable across the datasets