Tasks could have more logical structure than a dependency graph. You can imagine a kind of programming language for tasks with conditional logic, loops, exceptions, blocking I/O, etc. I don’t know if this has been explored?
I don't know. But thinking about it, writing code with complex structure usually makes sense only when it's executed more than once. Which leads me to think that the best place to look for insights on representing tasks are code isn't task management, where everything is unique and one-off, but process management, where a whole graph of tasks is executed repeatedly.