Would internal multi-threaded decoding be welcome? #695
Description
We maintain a fork (zenjxl-decoder) that adds rayon-based parallel decoding — parallel LF/HF group decode and parallel render pipeline, behind an optional threads feature flag.
On a 4K VarDCT image with 8 threads it's roughly 4-5x faster than single-threaded decode. The implementation is ~20 commits touching frame decode, render pipeline, and group scheduling.
Before we invest in cleaning this up for a PR, is internal multi-threading something you'd want in jxl-rs? For what it's worth, libdav1d (used for AVIF in Chromium) does use internal multi-threading for single-frame decode.
Totally understand if single-threaded is the intended scope — just want to check before proposing anything.