mps cross-attention hack: don't crash on fp16 (#2258)

* mps cross-attention hack: don't crash on fp16

* Make conversion explicit.

src/diffusers/models/cross_attention.py

@@ -251,7 +251,7 @@ class CrossAttention(nn.Module):
                 # HACK: MPS: Does not support padding by greater than dimension of input tensor.
                 # Instead, we can manually construct the padding tensor.
                 padding_shape = (attention_mask.shape[0], attention_mask.shape[1], target_length)
-                padding = torch.zeros(padding_shape, device=attention_mask.device)
+                padding = torch.zeros(padding_shape, dtype=attention_mask.dtype, device=attention_mask.device)
                 attention_mask = torch.concat([attention_mask, padding], dim=2)
             else:
                 attention_mask = F.pad(attention_mask, (0, target_length), value=0.0)