The test takes a lottie file and a timestamp in seconds and produces a
rendernode at that timestamp.
It then serializes that node and compares it via diff(1) with a file
containing the expected output.
This is a lot stricter than it needs to be (because different node files
can generate the same output and updates to the rendering pipeline can
break everything) but I chose this method on purpose because it does a
good job at guarding against accidental changes in other parts of the
code.
It's also better than comparing image output because it avoids
antialiasing artifacts when using curves and things like that.
e235392894