Adds support for CUDA accelerated vmaf, see: #163, #178, #320

Changes:

• I added both a --cuda argument, which enables the feature and a --cuda-hwaccel argument that allows one to disable hardware decoding which would otherwise be used.
  The rationale is that hardware decoders may not support the target codec; as an example my GTX 1660s cannot decode av1 media.
• I added some tests to ensure that the filter chains get built correctly when using the new flags. I also manually tested multiple combinations across different files and everything seems to work as expected.

Notes:

• I thought it was cleaner to extend the logic inside the ffmpeg_lavfi function instead of creating a new one. The logic remains the same and only a few strings differ, this avoids unnecessary duplication.
• When cuda is set to true, scale_cuda is always presesnt among the filters, even when scaling is not needed. That's to force conversion to yuv420p as it's the only format that works with vmaf_cuda, as tested in #163