Add Python image evaluation library callable via cpython bindings #22

Background

Image evaluation should be implemented as a clean Python library
callable directly from Go via the existing cpython bindings,
rather than as a subprocess bridge.

Proposed Approach

Create image_eval.py with a simple interface following the same
pattern as modeling/model.go:

def compare(reference_path, output_path):
# Returns dict with ssim, mse, psnr, is_grayscale, color_type

def passed(reference_path, output_path, threshold=0.80):
# Returns True if output passes quality threshold

Questions

1. Is this the right interface for the cpython bindings to consume?
2. Should this live in go-mfp or in the image evaluation repo?
3. Are there any constraints on what Python types can be returned
   via the cpython bindings?