-   Switch to `argparse` and clean up imports in `commands.py`.
-   Remove `requires-optional.txt` and `test_requirements/*`
    (put dependencies in `pyproject.toml` instead).
-   Remove `black` and use `ruff` instead.
-   Add `--noformat` option to `python commands.py codegen` for experimenting.
-   Pass output directory around to control what is formatted and linted.
-   Add `python commands.py format` to only do formatting.
-   Add `python commands.py lint` to check code.
-   Reformat comments in `codegen/*.py`.
-   Use double-quoted strings for (most) code generation.
-   Incorporate pandas version pinning from #5219.

Note:

1.  Only the generated code is formatted and linted:
    -   `plotly/validators/**/*.py`
    -   `plotly/graph_objs/**/*.py`
    -   `plotly/graph_objects/__init__.py`

2.  The strings in the data used by code generation are (for example)
    `"'some_name'"` (i.e., have embedded single quotes). This PR does
    not try to fix that: instead, we rely on reformatting to turn all
    the single-quoted strings into double-quoted strings.