I did find some issues, and I would have written them up but I ran out of time last night lol.
One thing from this weekend was some awkwardness with the InlineTextContent API. I haven't made a reproducer yet but passing in a new map doesn't always trigger a relayout of the text. You'll see a bunch of comments where I append zero-width characters to the string to force relayout when the placeholders' sizes change.
The reason I ran into that was due to trying to hide the placeholder API altogether. The cheat I'm using now is to use the first composition to measure the inline contents using the Text's layout constraints, then immediately recalculating new Placeholders with the results. This isn't ideal, since there's a frame where they content is all zero-sized and you see it jump into place (it's even visible in that gif). I don't know if that's avoidable though since the content would need to finish measurement + layout before starting text layout, and text layout is required to measure the text node, so there's a circular dependency.
Something that does seem fixable though is that Density throws an exception if you try to convert an em TextUnit to pixels. This makes sense in most cases, since ems are relative to font metrics and don't have any meaning outside of a specific font. But when inline content is composed, it is in the context of a specific font. It seems like it should be possible for Compose to merge all the SpanStyles and TextStyles, determine the font, load the font metrics, and pass a special Density ambient to inline composables that had support for em conversions. That said, this is only really an issue if they need to measure things in ems, which is really only necessary if you're trying to calculate the Placeholder lazily as described above.