Raise AffineError from __rmul__

[sgillies]

See #32.

[sgillies]

After some thought, I think we do want __rmatmul__() and __rmul__() to allow for vector (transpose) @ matrix multiplication that yields another transposed vector. So, only for 2-tuples on the left side.

[mwtoews]

I think these right-hand multipliers introduce trouble, since matrix math is not commutative. The __rmul__ op is intended for vector-matrix multiplication, whereas affine transforms need matrix-matrix or matrix-vector multiplication which is already supported by __mul__.

Demo, cross-referencing with NumPy. Note that vectors for matmul are in the form (x, y, 1):

>>> import numpy as np
>>> from affine import Affine
>>> A = Affine.translation(1, 1) * Affine.scale(2, 3)
>>> print(A)
| 2.00, 0.00, 1.00|
| 0.00, 3.00, 1.00|
| 0.00, 0.00, 1.00|
>>> A * (1, 1)
(3.0, 4.0)
>>> (1, 1) * A
<stdin>:1: DeprecationWarning: Right multiplication will be prohibited in version 3.0
(3.0, 4.0)
>>> np.array(A) @ (1, 1, 1)
array([3., 4., 1.])
>>> (1, 1, 1) @ np.array(A)
array([2., 3., 3.])

The last row is the correct answer from a linear algebra perspective, but this doesn't make sense for affine transformations. So I think the idea to raising an error was a good one to follow through.