feat: constant memory dataset DownloadValidate

[nayib-jose-gloria]

Reason for Change

• https://czi.atlassian.net/browse/VC-1338

Changes

• replace chunked matrix traversal in extract_metadata with call to count_matrix_nonzero in cellxgene_schema that does the same thing
• When populating dataset citation, update dataset.uns in-place using h5py rather than re-opening and re-writing the entire anndata object.
• Use memory efficient read_h5ad call from cellxgene_schema in extract_metadata
• update Validate batch jobs to use constant memory
• delete ProcessDownload (more complex logic no longer required if we are using a constant memory allocation) and instead incorporate the URI -> S3 upload into the ProcessValidate step
• Update SFN infra to remove obviated steps (Download, RegisterJobDefinition, DeregisterJobDefinition)

Testing steps

• local testing with large datasets
• TODO: rdev testing,

Notes for Reviewer

• TODO: determine what constant memory should be set at, based on testing
• TODO: fix tests as needed

[github-actions]

Deployment Summary

• backend_url: https://pr-7398-backend.rdev.single-cell.czi.technology
• backend_de_url: https://pr-7398-backend-de.rdev.single-cell.czi.technology
• backend_wmg_url: https://pr-7398-backend-wmg.rdev.single-cell.czi.technology
• frontend_url: https://pr-7398-frontend.rdev.single-cell.czi.technology
• delete_db_task_definition_arn: arn:aws:ecs:us-west-2:***:task-definition/dp-rdev-pr-7398-deletion:1
• migrate_db_task_definition_arn: arn:aws:ecs:us-west-2:***:task-definition/dp-rdev-pr-7398-migration:1