The current compressor struggles to compress hard edges in images with fast transparency changes when the backing RGB color in the transparent region is totally different to the non-transparent color (e.g. black). This can cause block artifacts along the transparent edges in the image.

The usual fix for this is to extrude the non-transparent edge color in to the transparent region, minimizing the magnitude of the color diff that must be represented. The current compressor requires users to manually pre-extrude their textures prior to compression, which many DCC tools can do, but we can probably do something specifically format-aware based on e.g. choosing a current partition endpoint color or the average of them.

We could do this automatically when using the -a option, but should also add an explicit enable/disable flag so users can control whether this optimization is applied.