Either angles or cld don't work consistently

Adding zero shouldn't silently influence results in a quite unexpected way:
```
julia> round(Int, cld(0+5u"°", 1u"°"))
57

julia> round(Int, cld(5u"°", 1u"°"))
5
```
Either it should throw an error, or remain consistent. Not sure what part exactly is to blame here: angles handling or the `cld` method. Probably the former?

Reproduce the two Julia expressions comparing cld with and without the added zero, then inspect the angle-handling and cld paths to identify where their behavior diverges. Done means the zero case either raises an error or produces a result consistent with the equivalent expression, with regression coverage for both examples.

Written by the indexing model from the issue text.