Treating antipodal quaternions as equal breaks some operations

Currently, antipodal quaternions are treated as equal, i.e., q == -q. I can see the motivation for this -- they induce the same rotation -- but this leads to several issues:

• AFAIK, accessing the quaternion components is part of the public API. So it's confusing that two UnitQuaternions can be == but have different values of the field: q == -q but q.w != (-q).w.

• The exponential map is broken: https://github.com/JuliaGeometry/Rotations.jl/issues/128

• zero(UnitQuaternion) is not a valid unit quaternion. As described in https://github.com/JuliaGeometry/Rotations.jl/issues/30#issuecomment-662464903), there is an undocumented semantics of "unnormalized unit quaternion" which is used for some operations such as exp and log, but is not ultimately coherent. Better would be to have a separate type for infinitesimal unit quaternions, similar to what's being done in https://github.com/JuliaGeometry/Rotations.jl/issues/30. That type has a zero; UnitQuaternion should not have a zero.

Start by reviewing the UnitQuaternion equality, exponential and logarithm operations, and zero(UnitQuaternion), then read the linked issues about the exponential map and infinitesimal unit quaternions. Done means the quaternion equality semantics and zero-value behavior are coherent, with a separate type if needed for infinitesimal operations.