In case anyone runs across similar AMD GPU issues, the "disable_idle_d3=1" option in the second pastebin link (when used as a kernel parameter) should be prefixed as "vfio-pci.disable_idle_d3=1". Even then, it can still be hit-or-miss with some cards (it works most of the time for me, but some VM restarts still find that the GPU has entered D3 and won't come back without rebooting the host).