fix(background): rasterize the heropattern tile at its target scale

[LeadcodeDev]

Severity Medium, category quality. Location: crates/rustmotion/src/engine/render/background.rs:605

Impact

tile_w = def.width * cfg.scale is computed at line 581 and used only for the overscan margin; the pixmap at 605-606 is sized from the unscaled def.width/def.height and resvg renders with Transform::default() (identity). The magnification then happens on the raster, via the shader matrix, with SamplingOptions::default() — which I read in skia-safe-0.99.0/src/core/sampling_options.rs:70-80: filter: FilterMode::Nearest, mipmap: MipmapMode::None. So heropattern with scale: 4 on a 1080p frame draws hard 4x4 pixel blocks from a 20x20 source raster, when the source is vector art that resvg could have rendered crisply at any size. The whole point of shipping 87 SVG patterns is lost above scale 1. It is also wasteful below 1: a scale: 0.25 pattern is rasterized at full size and then point-sampled down, aliasing badly.

Fix

Size the pixmap at the scaled tile ((def.width * cfg.scale).ceil() x (def.height * cfg.scale).ceil(), clamped to a sane maximum) and pass tiny_skia::Transform::from_scale(cfg.scale, cfg.scale) to resvg::render, then drop the shader matrix so the tile is used 1:1. At minimum, pass SamplingOptions::new(FilterMode::Linear, MipmapMode::Linear) instead of the nearest-neighbour default.

Evidence the audit read

let pw = def.width.ceil() as u32;
    let ph = def.height.ceil() as u32;
    let Some(mut pixmap) = tiny_skia::Pixmap::new(pw, ph) else {
        return;
    };
    resvg::render(&tree, tiny_skia::Transform::default(), &mut pixmap.as_mut());
...
    let matrix = if cfg.scale != 1.0 {
        Some(skia_safe::Matrix::scale((cfg.scale, cfg.scale)))
    } else { None };
    let Some(shader) = tile_image.to_shader(
        (skia_safe::TileMode::Repeat, skia_safe::TileMode::Repeat),
        skia_safe::SamplingOptions::default(), matrix.as_ref(),

Stacked on fix/heropattern-colour-escape, which carries the previous finding of this workstream. GitHub shows only this finding's diff; merge in order.

Part of the September 2026 audit remediation chantier. Refs #220 (RM-22).