Treat convention as setting ignore, rather than select

[charliermarsh]

This PR treats the convention setting as an "ignore" directive, rather than a "select" directive, which leads to confusing settings and behavior.

The new semantics are such that setting a convention force-disables any rules that are incompatible with that convention. (It's effectively as-if those rules don't exist.) You can then disable further rules via ignore directive.

For example, this would now be a correct and reasonable configuration:

[tool.ruff]
select = [
    "D",
]
ignore = [
    "D1",
]

[tool.ruff.pydocstyle]
convention = "google"

The biggest limitation here is that you can't enable rules that aren't compatible with the convention, as long as the convention is set. It's impossible. So for example, if you used pep257 but wanted to enable a few of the checks that pep257 excludes by default, you'd be out of luck. This is a limitation, and I considered a few other models (like letting extend-select override this behavior)... but honestly, it felt like an edge case, and all the other models had their own confusing behaviors. If you find yourself in that position, you can always avoid setting convention and select codes manually.

See the discussion here: #1569. This approach also would've prevented this issue: #1603.

[charliermarsh]

\cc @stinodego - how does this look?

[stinodego]

Thanks @charliermarsh ! Exactly what I had in mind.

Regarding the limitation of this approach: I believe this now matches how flake8-docstrings is normally configured. So even though it is a limitation, it will at least be familiar to people coming from flake8. And it's easy to work around, by creating your 'own' convention as a set of ignores. This is what we previous did (and still do now with Ruff) in Polars.

[ManonMarchand]

Perfect, thanks!