Transform decomposition

[groutr]

A useful utility function for decomposing a conformal transform into its translation, rotation, and scale parameters.

from rasterio.transform import decompose
# transform = Affine.translation(1, 3) * Affine.scale(3, 3) * Affine.rotation(45)
decompose(transform)
# ((1.0, 3.0), 44.999999999999986, (3.0, 3.0))

[snowman2]

I am wondering if this makes more sense here?
https://github.com/rasterio/affine/

[groutr]

This PR could make more sense for affine if the following two things can be resolved.

1. In its current state this would add a numpy dependency to affine (which affine seems to be trying to avoid external dependencies). Affine seems to do a similar decomposition, but only depends on the standard library. I'm not sure how robust affine's calculation is regarding floating point issues.
2. Affine does have methods to extract translation, rotation (rotation_angle), and scale (scale_factor) but will return those values whether the transform is conformal or not. This implementation limits decomposition to conformal transforms. Checking for conformity in affine could potentially break compatibility in affine. It could be added as a new method perhaps where conformity is checked.