Fix fusion of convolution and batch normalization layers not preserving dtype

[wico-silva]

Fixes the error:

Traceback (most recent call last):
  File "tools/demo.py", line 320, in <module>
    main(exp, args)
  File "tools/demo.py", line 290, in main
    model = fuse_model(model)
  File "YOLOX/yolox/utils/model_utils.py", line 92, in fuse_model
    m.conv = fuse_conv_and_bn(m.conv, m.bn)  # update conv
  File "YOLOX/yolox/utils/model_utils.py", line 74, in fuse_conv_and_bn
    fusedconv.bias.copy_(torch.mm(w_bn, b_conv.reshape(-1, 1)).reshape(-1) + b_bn)
RuntimeError: expected scalar type Half but found Float

when using --fp16 --fuse. To replicate the issue, run the command:

python tools/demo.py image -n yolox-tiny -c yolox_tiny.pth --device gpu --conf 0.01 --fp16 --fuse