Add unit tests for the plotting functions; fix scales >= 1.3.0 compatibility

[zeehio]

Summary

None of this package's plot functions render to a graphics device — they build and return a ggplot object (or, for overlay_peaklist(), a list of ggplot2 layers) via + composition. That means they can be unit-tested like any other data structure, by inspecting the returned object's data/labels/scales/layer classes, without snapshot/visual-regression tooling (e.g. vdiffr).

• mat_to_nativeRaster() (utils-plot.R): pure matrix→raster encoding — shape/class/channels attributes, determinism, character vs. pre-encoded colormap, and the zero-length edge case
• plot_interactive(): wraps a ggplot in plotly::ggplotly(), checked it returns a "plotly" object
• cubic_root_trans(): transform/inverse round-trip (including negative values, since it's a signed cube root) and the breaks() empty/non-finite edge case
• as.data.frame.GCIMSSample(): melts the intensity matrix, checked columns/values, dt_range/row.names
• plot() on GCIMSSample/GCIMSChromatogram/GCIMSSpectrum: labels, data, axis-range restriction, the remove_baseline branch, and the single-value/multi-value/empty-value subtitle branches
• overlay_peaklist(): the color_by (none/literal color/column), apex, >10-groups-hides-legend, palette-recycling, non-data.frame input, and pdata-merging branches, plus its two validation errors

Brings plot-GCIMSSample.R, plot-GCIMSChromatogram.R, plot-GCIMSSpectrum.R, and utils-plot.R to 100% line coverage. Overall package coverage: 78.34% → 86.83%.

Bug fix

While testing plot(sample, trans = ...), found that mat_to_gplot() checked inherits(trans, "trans") to validate a transform object passed directly (not as a string). scales >= 1.3.0 renamed this S3 class from "trans" to "transform", so passing any transform object this way — including the package's own cubic_root_trans() — always failed with "unknown trans value", even though the roxygen docs explicitly point to this as valid usage (plot(s, trans = scales::identity_trans()), plot(s, trans = cubic_root_trans())).

Fixed the check to use the class name scales has used since 1.3.0 (already over a year old), and bumped DESCRIPTION's scales dependency to >= 1.4.0 accordingly. The string-based trans = "log10" path was unaffected (it goes through a different code path using the still-preserved legacy *_trans() function names).

Test plan

• Ran each new test file individually — all pass
• Full suite testthat::test_local(".") — 672 passing, 1 pre-existing skip, 0 failures
• covr::package_coverage() confirms all four files at 100%; overall package coverage 78.34% → 86.83%
• Manually verified the trans= bug end-to-end before and after the fix (string values, transform objects, and the two error paths)

---

Generated by Claude Code

[zeehio]

On this sprint we've gone from testing 23% of the code lines to 86% of the code lines.

A higher test coverage helps to understand if something is breaking because of a dependency upgrade or because an unintended regression in a recent code change. It also gives confidence when making a change to see that the change we are making doesn't break the code we have so far. It also helps detecting issues in less visited code paths, such as error handling or corner features.

During this process we have fixed a dozen bugs.

One of these bugs prevents spurious peaks from appearing in the peak table, before I remember we were filtering them out manually.

Another bug fixed the saturation column in the peak lists. This column was not used in downstream analysis, but still it is a useful QC metric.

The other 10 fixed bugs would not be triggered usually, but still it's worth having them fixed.