Should we expose the direction in struct fcft_text_run?
Since you can feed any kind of mixed LTR/RTL strings to fcft_text_run_rasterize(), and fcft will typically be able to emit something that appears sane, you can technically have both LTR and RTL in the same text run.
So, should we expose a single direction for the entire run? Which one? The first one detect?
Or do we add an array? Kind of like the cluster array. Here, we'd map cluster ranges to LTR/RTL.