This is faster, and allows using 2-core instances instead of
requiring a minimum of 8 GB / 4 cores.

[skip ci]

This PR moves our repo setup to onCreateCommand from postCreateCommand in order to resolve a race condition between conda init and the default shell in VS Code. You can test this new devcontainer.json by creating a codespace for my fork/branch: https://github.com/codespaces/new?hide_repo_select=true&ref=jungaretti%2Fupdate-devcontainer&repo=576410652

postCreateCommand doesn't block interaction with your codespace. You can use other terminals and move around editors while it runs. On the other hand, onCreateCommand runs before you interact with your codespace. You won't be able to open a terminal until it finishes. onCreateCommand is a better choice for running our setup script because we need to run conda init before using the terminal.

This PR fixes numpy's devcontainer.json to successfully create an Anaconda environment. You can test this new devcontainer.json by creating a codespace for my fork/branch: https://github.com/codespaces/new?hide_repo_select=true&ref=jungaretti%2Ffix-devcontainer&repo=576410652

Changes

* Removed references to Microsoft's anaconda-0.3 image. This image is deprecated. For a list of maintained images, see https://github.com/devcontainers/images
* Added a new setup.sh script that is run by postCreateCommand and removed the unnecessary Dockerfile. For more information about devcontainer lifecycle scripts, see https://containers.dev/implementors/json_reference/#lifecycle-scripts
* Increases memory requirement to 8 GB. Anaconda environment creation seems to fail when only 4 GB of memory are available. This hostRequirements addition will ensure that the environment is created successfully with GitHub Codespaces.
* Remove unnecessary rust dependency

* initial files

* STY: remove noop.txt (we have an environment.yaml)

some other style fixups

Co-authored-by: Sarah Kaiser <sckaiser@sckaiser.com>