run mypy via pre-commit

[jameslamb]

Contributes to #115.

Proposes running mypy via pre-commit, for the following reasons:

• reduce duplication of mypy configuration
• remove a dependency from the [test] extra for the package
• standardizes mypy version across all the place where those checks are run (to limit "it works on my computer, why is CI broken" types of situations)

As part of this, this PR also proposes running pre-commit checks in CI before the other building and testing, on a standard GitHub-hosted ubuntu-latest runner, to avoid occupying an NVIDIA-hosted GPU runner for a build that's doomed to fail because of linter errors.

Notes for Reviewers

Running these checks in the rapidsai/ci-conda:latest image and via a script named ci/check_style.sh mirrors what is done across most RAPIDS libraries. That'll make it easier to continue integrating CI tools from RAPIDS.

[RAMitchell]

@trivialfis did this for a reason - maybe he can explain why.

[trivialfis]

The pre-commit hook runs in an isolated environment, which doesn't include legate and other dependencies. As a result, the type checking is not rigorous.

[jameslamb]

The pre-commit hook runs in an isolated environment, which doesn't include legate and other dependencies. As a result, the type checking is not rigorous.

Blegh you're right, I misunderstood how the isolation for mypy worked. We could get pretty good coverage of most dependencies by adding them to the additional_dependencies list that pre-commit provides... but not legate-core or cunumeric, because they don't publish wheels today.

We can close this, sorry for the noise and thanks for correcting my understanding.

[RAMitchell]

Just had a quick read about this issue: https://tech.quantco.com/blog/conda-in-pre-commit

Looks like it is possible now with (too much) effort to use a conda environment with pre-commit. Maybe in the future this will be easier.

[jameslamb]

ooo very interesting, hadn't seen that! Thanks for sharing.