Avoid typechecking scalar arguments in transform.xy/rowcol

[groutr]

Resolves #2283

[sgillies]

@groutr I tried a different approach in #2285, checking for __iter__ is the most general test and this aspect of being a collection is the only thing we require (we need the inputs to be zip-pable and that requires __iter__).

[groutr]

@sgillies Checking isinstance(x, Iterable) is equivalent to checking for __iter__ (https://docs.python.org/3/library/collections.abc.html#collections.abc.Iterable)

In this PR, if scalar arguments are passed, I wrap them in an iterable container before applying the transformation. That makes the inputs zip-able. After the transform, I return based on if the inputs were scalar or not.

Another important part of this PR is the chained exception, preserving the TypeError that is raised from Affine. That TypeError provides important type information if the multiply failed.

[sgillies]

@groutr I appreciate your work and don't dismiss this PR lightly! I only want to add the minimum of new code and behavior and am not entirely sure that everything we want to support subclasses Iterable.

raise from would be appropriate for 1.3.0 for sure and since we're being more selective about numpy versions there I think testing Iterable would be fine too.