Add image comparison tests

[ecomodeller]

Problem

Currently, plot tests mainly verify that functions return objects, not their visual correctness. There's only one image comparison test (for wind rose) which uses
pixel-by-pixel comparison and is acknowledged as fragile in a TODO comment.

This means refactoring plotting code could introduce visual regressions that go undetected.

Proposal

Add visual regression testing using pytest-mpl, the standard tool for matplotlib image comparison in scientific Python.

Why pytest-mpl over alternatives like ApprovalTests?

Aspect	pytest-mpl	ApprovalTests
Matplotlib integration	Purpose-built	Generic, manual setup
Tolerance support	Built-in RMS tolerance	Basic
Cross-platform	Handles rendering differences	More fragile
Image diffs	Automatic on failure	Manual

Example test

@pytest.mark.mpl_image_compare(tolerance=2)
def test_scatter_plot(comparer):
    fig, ax = plt.subplots()
    comparer.plot.scatter(ax=ax)
    return fig

Plot types to cover

• timeseries
• scatter
• wind rose (replace existing fragile test)
• taylor diagrams

Other plot types are less crucial.

Considerations

• Cross-platform: Need tolerance-based comparison since fonts/rendering differ between OS
• Plotly backend: Would need a separate approach (pytest-mpl is matplotlib-only)
• Baseline management: pytest-mpl generates baselines with --mpl-generate-path