Excessive memory consumption when converting some images

Hi,
I have an issue where converting some particular images takes up excessive amounts or memory and time.

[memory_svg.zip](https://github.com/Kozea/CairoSVG/files/2533488/memory_svg.zip)

This image takes 50 seconds and 3Gb of memory to successfully convert.
What I have observed is some kind of threshold where images above 1Mb will sometime take exponential amount of memory (up to > 50Gb for 6Mb images).

I am using cairosvg 2.2.1 with python 3.6 from official docker image.

Start by reproducing the conversion with the attached memory_svg.zip using the reported CairoSVG 2.2.1 and Python 3.6 environment, then profile memory and runtime as image size changes. Done means the attached image and similarly large images convert without the reported excessive memory consumption or extreme delay.

Written by the indexing model from the issue text.