radv: implement DCC fast clears using comp-to-single on GFX10+
comp-to-single allows to skip FCE. This will allow us to enable fast clears for concurrent images in a separate change later.
No CTS regressions with RADV_DEBUG=forcecompress on Sienna Cichlid and Navi10.
Based on !10502 (merged) and !10516 (merged)