Combine image soft masks of a different size at the larger resolution

[Velli20]

Summary

An image's /SMask may have different dimensions from the image itself, since both map onto the unit square. The decoder paired mask and image pixels one to one. A low-resolution color image therefore kept its own size and took its alpha from only the first few mask samples.

Some producers draw each line of text as a 2×2 single-color image whose full-resolution soft mask holds the glyph shapes. Those lines rendered fully transparent, so the page appeared to have no text.

When the sizes differ, the decoder now:

• enlarges whichever of the image and mask is smaller by nearest-neighbour sampling, to the larger width and the larger height;
• combines them with the existing alpha path.

When the sizes already match, nothing changes.

Changes

• pdf-image/src/raster.rs: resample_nearest scales a Gray8 or RGBA image to a target size. It tolerates short decoder buffers and returns a resource-limit error if the size overflows or allocation fails.
• pdf-image/src/image_decoder.rs: apply_resized_soft_mask handles mismatched sizes. The existing assembly code moved into render_ready_image.
• pdf-image/src/error.rs: PdfImageError::Raster wraps ImageRasterError.

Test plan

• cargo conformance render on an affected document: text lines that were missing now render in their colors.
• cargo test -p pdf-image -p pdf-graphics -p pdf-resources -p pdf-canvas, cargo clippy --all --workspace, cargo fmt --check.

🤖 Generated with Claude Code

[github-actions]

Conformance reads

No document reads differently from the base commit any more.

840 documents of the PDFium corpus read as on 5fc63f6.
1008 documents of the pdf.js corpus read as on 5fc63f6.

Workflow run