radv: Fix interpolation for non-float types
Found this issue while trying to get the text in the overlay to work.
The second patch fixes the overlay, the first one avoids CTS regressions while doing so.
(Also the CTS completely has no tests that fp16 interpolation work even though the spec allows it)
"Fragment shader inputs that are signed or unsigned integers, integer vectors, or any double-precision floating-point type must be decorated with Flat."