Investigate the typical magnitude of floating-point noise in `rasterio` cell dimensions

[b-j-karl]

I think we need to investigate the typical magnitude of floating-point noise to find an appropriate tolerance for np.isclose. Geographic coordinate systems for example may have small differences in cell width and height but at a larger scale these small differences can add up to a substantial sum.

Originally posted by @b-j-karl in #413 (comment)