Thanks for the hint.
This is a bug. The cbrt function is supposed to work with complex numbers. I am pretty sure this is a regression, it must have worked at some point before.
Please submit an issue on the bug tracker.
From the documentation:
cbrt(x)
Compute the third (cubic) root of x. In real mode, it accepts any real number. Negative numbers will yield a negative cubic root.
[...]
In complex mode, this function accepts any complex input. The result will generally be the first complex root, i.e. the one with a phase between 0 and π/3. Real negative arguments however will still yield a real (negative) result, matching the function’s behavior in real mode. Use x^(1/3) to get the first complex root.
True to the mathematic definition of a function, it can only ever return a single value. It can not return all roots at one.