Use single twiddle factor table across all powers of two

The generation of a single twiddle factor table is as follows:

```python
sign = -1 if not inverse else 1
k = np.arange(n, dtype=jnp.float64)
return np.exp(1j * sign * 2 * np.pi * k / n)
```

The strided access pattern for bit-reversed indexing is as follows;

```python
step = n // size
half = size // 2
for k in range(half):
    W[k * step] 
```