Fix stable diffusion output error on WebGPU (tinygrad#2032)

* Fix stable diffusion on WebGPU

* Remove hack, numpy cast only on webgpu

* No-copy numpy cast

examples/stable_diffusion.py

@@ -651,12 +651,14 @@ def do_step(latent, timestep, index):
 
  # make image correct size and scale
  x = (x + 1.0) / 2.0
- x = (x.reshape(3,512,512).permute(1,2,0).clip(0,1)*255).cast(dtypes.uint8)
+ x = (x.reshape(3,512,512).permute(1,2,0).clip(0,1)*255)
+ if Device.DEFAULT != "WEBGPU": x = x.cast(dtypes.uint8)
  print(x.shape)
 
  # save image
  from PIL import Image
- im = Image.fromarray(x.numpy())
+ import numpy as np
+ im = Image.fromarray(x.numpy().astype(np.uint8, copy=False))
  print(f"saving {args.out}")
  im.save(args.out)
  # Open image.