I also have a background in Olympic weightlifting and know exactly how convoluted complexes can get.
Currently, I model those as either supersets (when it fits) or I name the complex as the movement and then use the metadata to get at a data shape that mostly fits.
Example of supersets would be something like
Clean: 150 2r 5s + Jerk: 150 5s
For a clean double, single jerk. The named complex just captures the reps and weights per submovment in the metadata.
It feels like using supersets for these would make tracking data over time difficult. eg given your clean clean jerk example, if one wants to track volume of a particular exercise it's not clear what the volume for each of those should be applied to, and it could get rather messy (is it 2x cleans + a jerk, or 1x clean + a clean and jerk?). Everyone probably has different preferences there, but my point is that native support for complexes would probably make stuff like this easier by providing a way to properly define/separate each component of the performance at data entry time.
I don't mean to only talk about things it doesn't support, though. I honestly think this looks great even without support for complexes (it's just long been the top of my wishlist for exercise tracking tools!) and am going to try it out next week!
I will do some noodling on this. If you have ideas, I’d love to dialog on them. I hadn’t implemented anything beyond the basics just because I hadn’t been doing them lately. I definitely want to support your use case as a first class language feature.