A data-mapped scalar aesthetic on boxplot / violin is dropped by the stat, then fails validation

Summary

The boxplot / violin stat drops non-grouping aesthetic columns from its output, so a mapped scalar aesthetic fails validation after the stat has run, with an error that names an internal column list. Grouping aesthetics (fill, stroke) survive because they partition the summary; scalar ones (linewidth, size, linetype, opacity) do not.

Reproduction

VISUALISE species AS x, body_mass AS y, bill_len AS linewidth FROM ggsql:penguins
DRAW boxplot

Failed to generate Vega-Lite output: Validation error: Column 'linewidth' referenced in
aesthetic 'linewidth' (layer 1 (global data)) does not exist.
Available columns: __ggsql_aes_pos1__, __ggsql_aes_type__, __ggsql_aes_pos2__, and __ggsql_aes_pos2end__

Control — the same query with sex AS fill renders fine.

Expected

Either carry a scalar aesthetic through the stat (it is constant within a group whenever it is part of the grouping, and otherwise needs a documented aggregate), or fail with an error that names the actual constraint ("aesthetic linewidth cannot be mapped on a boxplot layer") rather than leaking __ggsql_aes_* columns to the user.

Start with the boxplot and violin stat, then trace how scalar aesthetics and grouping aesthetics are passed into post-stat validation. Reproduce the penguins query and compare it with the sex AS fill control. Done means scalar aesthetics no longer fail after the stat, or the error clearly states the supported constraint without exposing internal columns.