Feature: Add handler for fft primitive

[bdrhill]

Description

The fft primitive is not implemented. This affects jnp.fft.fft, jnp.fft.ifft, and related FFT operations.

MWE

import jax.numpy as jnp
import asdex

def f(x):
    return jnp.fft.fft(x).real

x = jnp.array([1.0, 2.0, 3.0, 4.0])
J = asdex.jacobian(f, x, output_format="dense")(x)  # NotImplementedError

Error

NotImplementedError: No handler for primitive 'fft'.

Implementation Notes

The FFT is a linear operation with a dense Jacobian (each output depends on all inputs).
The sparsity pattern is fully dense for 1D FFT, but for multi-dimensional FFTs with axis selection,
the pattern has block structure matching the FFT axes.