Add a note to use --force-fp32 on some amd cards

I was struggling to figure out why svd was super slow on my amd card (took about half an hour to generate 1 video), and it ended up being the lack of the --force-fp32 flag (now takes 1 minute). Could you please indicate this somewhere in the amd run guide in the readme for people in the future who might have a similar problem? Btw I am using an instinct mi25 flashed to a wx9100 (which is a card that is heavily fp32 compute).