v3d: Add bias to fix depth buffer rounding

The floating point to fixed point conversion for depth values floors the value instead of rounding it as required by the OpenGL spec. For fixed point formats add a bias of 0.5 to causing rounding behavior when the value is floored.