Consider ditching codacy

[ajjackson]

Codacy can be a bit of a nuisance as it involves an external service, API keys etc. This leads to e.g. "failing" CI for branches from external contributors.

Instead we could be running ruff and/or pylint in a Github action. The output is not quite as pretty, but simpler and more reliable to access.

The main thing it provides (via further complexity/failure-points) is a nice interface for the code coverage results. For testing we already have a nice little action/bot setup that writes high-level test results to a comment in the PR. Maybe we can do something similar for coverage data, and rely on local analysis for fancy HTML views?