From ggsql: Bad plot background dimensions

A polar plot reports desired_panel_aspect, which build_composition_grid turns into patch.aspect(1, 1) (src/plot/composition.rs:360-373). That aspect-locks the patch. Measured across canvas sizes:

canvas W | patch background x-range | legend text ends -- | -- | -- 960 | 0 … 960 | 905.9 1152 | 0 … 1152 | 1097.9 1344 | 78.6 … 1265.4 | 1289.9 1536 | 174.6 … 1361.4 | 1481.9 1920 | 366.6 … 1553.4 | 1865.9

The background keeps a constant width (1186.75) and is centred; the panel moves with it (+96 for a +192 canvas). The legend, though, is pinned to the full canvas — it always ends at canvas_width − 54.12, shifting by the full +192. So once the canvas is wide enough for the aspect lock to bite, the legend walks out of the patch it belongs to. Cartesian SETTING ratio => 1 takes the AspectMode::Range path instead and keeps a full-canvas background, which is why only polar (and presumably maps) show it.

The opaque canvas hides the symptom completely, since the legend is still drawn correctly — but the underlying layout inconsistency is still there and would show up again anywhere the patch background is not the canvas colour.

A polar plot reports desired_panel_aspect, which [build_composition_grid](https://github.com/posit-dev/hephaestus) turns into patch.aspect(1, 1) (src/plot/composition.rs:360-373). That aspect-locks the patch. Measured across canvas sizes:

canvas W patch background x-range legend text ends
960 0 … 960 905.9
1152 0 … 1152 1097.9
1344 78.6 … 1265.4 1289.9
1536 174.6 … 1361.4 1481.9
1920 366.6 … 1553.4 1865.9
The background keeps a constant width (1186.75) and is centred; the panel moves with it (+96 for a +192 canvas). The legend, though, is pinned to the full canvas — it always ends at canvas_width − 54.12, shifting by the full +192. So once the canvas is wide enough for the aspect lock to bite, the legend walks out of the patch it belongs to. Cartesian SETTING ratio => 1 takes the AspectMode::Range path instead and keeps a full-canvas background, which is why only polar (and presumably maps) show it.

The opaque canvas hides the symptom completely, since the legend is still drawn correctly — but the underlying layout inconsistency is still there and would show up again anywhere the patch background is not the canvas colour.

Start in src/plot/composition.rs:360-373 and trace how desired_panel_aspect becomes patch.aspect(1, 1) in build_composition_grid. Compare the polar layout with the Cartesian AspectMode::Range path across the canvas widths listed in the issue. Done means the patch background and legend remain aligned at wide canvas sizes without regressing the Cartesian behavior.