Transforms should implement an `ndim` property

[oesteban]

Something like:

>>> from nitransforms import linear
>>> linear.Affine().ndim
3

(only one timepoint)

>>> from nitransforms import linear
>>> linear.LinearTransformsMapping().ndim
4

In the future, we can imagine two more use cases:

>>> from nitransforms import linear
>>> linear.ComplexAffine().ndim
4

(only one timepoint)

>>> from nitransforms import linear
>>> linear.ComplexLinearTransformsMapping().ndim
5

cc/ @jmarabotto

[oesteban]

We just realized they do. So this is going to be changed a little:

• TransformBase -> should raise a TypeError instead of returning the dimensionality of the reference.
• Affine -> should overshadow TransformBase's property and return self._matrix.ndim + 1. That will also work for LinearTransformArray without loss of generality.
• Nonlinear transforms -> for the moment these cannot be 3D so they should redefine the property and return self._reference.ndim.