| <!-- | |
| Keep this short: reviewers can read the diff, so cover only what it cannot. | |
| Include only what applies, and delete these comments and any line you leave empty. | |
| See "Writing PR descriptions" in AGENTS.md: https://github.com/linebender/vello/blob/main/AGENTS.md#writing-pr-descriptions | |
| - Why: the problem or motivation, with links to issues (e.g. "Fixes #123"). | |
| - How it works, if the approach is not obvious from the code. | |
| - Where to look, if most of the diff is mechanical. | |
| - Risks: edge cases, behavior changes, and which renderers are affected | |
| (vello_cpu, vello_gpu wgpu, vello_gpu WebGL, research). | |
| - Performance: benchmark results for performance-sensitive changes, including regressions. | |
| - Testing: what you ran, and which snapshots you added or updated. | |
| --> | |
| <!-- | |
| Before requesting review: | |
| - Changes visible in the rendered output have a `#[vello_test]` snapshot test in `vello_tests/tests/`. | |
| For a bug fix, commit the test first so the reference image shows the bug, then the fix. | |
| - Mechanical changes (renames, moves, formatting) and snapshot updates are in their own commits. | |
| - User-facing changes to published crates have an entry under `[Unreleased]` in that crate's CHANGELOG.md. | |
| --> | |
| <!-- | |
| If an LLM generated any of this PR (code, tests, or this description), say so, for example: | |
| "Created with assistance from [model name and version]." | |
| See the Linebender LLM policy: https://linebender.org/wiki/llm-policy/ | |
| --> |