BFloat16 type support needed

Here is a minimal example demonstrating the issue:
```
import torch
import pywt

# Create a tensor with bfloat16 type
x = torch.randn(1, 1, 10, 10, 10, dtype=torch.bfloat16)

# Attempt to perform wavelet decomposition
coeffs = pywt.wavedec(x.numpy(), 'bior2.2', level=1, mode='symmetric')
```