CUDA: use FP32 arithmetic for conv2d

[JohannesGaessler]

Fixes #15680 .

The kernel can simply be made to always use FP32 arithmetic, this also has the advantage of not gimping performance on Pascal. Performance differences (on an RTX 4090) are negligible:

Backend	GGML op	Op parameters	TFLOPS ef47691	TFLOPS 2721aaf	Speedup
CUDA0	CONV_2D	ne_input=[16,16,128,8],ne_kernel=[3,3,128,512],type_kernel=f16,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	1.93	1.94	1.00
CUDA0	CONV_2D	ne_input=[16,16,128,8],ne_kernel=[3,3,128,512],type_kernel=f32,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	2.06	2.06	1.00
CUDA0	CONV_2D	ne_input=[19,19,256,16],ne_kernel=[4,4,256,4096],type_kernel=f16,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	2.26	2.33	1.03
CUDA0	CONV_2D	ne_input=[19,19,256,16],ne_kernel=[4,4,256,4096],type_kernel=f32,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	2.36	2.37	1.00
CUDA0	CONV_2D	ne_input=[19,19,4,16],ne_kernel=[2,2,4,4],type_kernel=f16,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	0.16	0.16	0.99
CUDA0	CONV_2D	ne_input=[19,19,4,16],ne_kernel=[2,2,4,4],type_kernel=f32,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	0.16	0.16	1.00
CUDA0	CONV_2D	ne_input=[19,19,8,16],ne_kernel=[4,4,8,128],type_kernel=f16,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	1.87	1.93	1.03
CUDA0	CONV_2D	ne_input=[19,19,8,16],ne_kernel=[4,4,8,128],type_kernel=f32,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	1.94	1.94	1.00
CUDA0	CONV_2D	ne_input=[19,19,8,16],ne_kernel=[4,4,8,130],type_kernel=f16,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	1.81	1.86	1.03
CUDA0	CONV_2D	ne_input=[19,19,8,16],ne_kernel=[4,4,8,130],type_kernel=f32,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	1.88	1.88	1.00
CUDA0	CONV_2D	ne_input=[224,224,1,1],ne_kernel=[2,2,1,8],type_kernel=f16,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	0.26	0.26	1.00
CUDA0	CONV_2D	ne_input=[224,224,1,1],ne_kernel=[2,2,1,8],type_kernel=f32,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	0.27	0.27	1.00
CUDA0	CONV_2D	ne_input=[224,224,1,8],ne_kernel=[2,2,1,8],type_kernel=f16,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	0.31	0.32	1.00
CUDA0	CONV_2D	ne_input=[224,224,1,8],ne_kernel=[2,2,1,8],type_kernel=f32,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	0.32	0.32	1.00
CUDA0	CONV_2D	ne_input=[224,224,3,1],ne_kernel=[3,3,3,8],type_kernel=f16,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	1.05	1.05	1.01
CUDA0	CONV_2D	ne_input=[224,224,3,1],ne_kernel=[3,3,3,8],type_kernel=f32,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	1.09	1.09	1.00
CUDA0	CONV_2D	ne_input=[58,58,32,1],ne_kernel=[3,3,32,64],type_kernel=f16,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	1.49	1.49	1.00
CUDA0	CONV_2D	ne_input=[58,58,32,1],ne_kernel=[3,3,32,64],type_kernel=f32,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	1.57	1.57	1.00
CUDA0	CONV_2D	ne_input=[58,58,32,8],ne_kernel=[3,3,32,64],type_kernel=f16,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	1.89	1.90	1.01
CUDA0	CONV_2D	ne_input=[58,58,32,8],ne_kernel=[3,3,32,64],type_kernel=f32,stride0=1,stride1=1,padding0=0,padding1=0,dilation0=1,dilation1=1,cwhn=0	2.00	2.00	1.00