Make px.sunburst/treemap/icicle sector order deterministic for Polars input

[Irahan2]

Link to issue

Closes #5765

Description of change

process_dataframe_hierarchy now adds a temporary row-index column, takes its minimum in each group, and sorts every level of the hierarchy by it after the group_by. Sectors are therefore always listed in order of first appearance for every dataframe backend. pandas and PyArrow already produced that order; Polars, whose group_by does not keep row order, produced a different order on every run.

Demo

Before, three runs of the script from the issue with a Polars DataFrame:

['West/Tech', 'West/Finance', 'North/Finance', 'South/Finance', 'North/Tech', 'South/Tech', 'South', 'North', 'West']
['South/Tech', 'West/Tech', 'North/Tech', 'South/Finance', 'North/Finance', 'West/Finance', 'West', 'North', 'South']
['South/Tech', 'West/Tech', 'North/Tech', 'North/Finance', 'South/Finance', 'West/Finance', 'West', 'South', 'North']

After, every run (and pandas / PyArrow):

['South/Tech', 'North/Finance', 'South/Finance', 'West/Tech', 'North/Tech', 'West/Finance', 'South', 'North', 'West']

Testing strategy

Added test_sunburst_treemap_with_path_order, parametrized over all the dataframe constructors and over px.sunburst, px.treemap and px.icicle. It checks the exact order of ids and values. It fails for the Polars constructor on main and passes with this change. The rest of tests/test_optional/test_px passes locally, apart from 3 tests that fail the same way on unmodified main in my Windows environment (test_timezones and test_date_in_hover for the PyArrow constructor, and one test_imshow_source_dtype_zmax case). The hierarchy tests also pass with the minimum supported narwhals (1.15.1).

Additional information (optional)

• pandas output does not change: it already used first-appearance order (narwhals' group_by uses sort=False). This matches how Plotly Express orders categories elsewhere (unique(maintain_order=True)).
• Only the hierarchy code path is affected. The other group_by in _core.py is only used for key lookups, and trace order there already comes from unique(maintain_order=True).
• AI disclosure: I used Claude Code (Claude Opus 5) to help find this bug (by comparing Plotly Express output for pandas, Polars and PyArrow inputs) and to draft the fix and the test. I have reviewed the changes and run the tests myself.

Guidelines

• I have reviewed the pull request guidelines and the Code of Conduct and confirm that this PR follows them.
• I have added an entry to the changelog if needed (not required for documentation PRs).