[Feature][Relax][ONNX] Support DFT (Discrete Fourier Transform) in Relax ONNX frontend #20960

Description

The ONNX DFT operator (opset 17 and opset 20) is currently missing from the Relax ONNX frontend (python/tvm/relax/frontend/onnx/onnx_frontend.py), as noted in RFC tracking issue #20124 under signal processing operators.

The DFT operator computes the Discrete Fourier Transform or Inverse Discrete Fourier Transform along a given axis for real ([..., 1]) or complex ([..., 2]) inputs.

Key Requirements

• Support forward (inverse=0) and inverse (inverse=1) DFT.
• Support both full spectrum (onesided=0) and one-sided spectrum (onesided=1).
• Support axis as an attribute (opset 17) and axis as an input tensor (opset 20).
• Support optional dft_length input for zero-padding or truncating the signal along the specified axis.
• Arbitrary input dimensions and transformation along any valid axis.
• Numerical correctness verified against ONNX Runtime and numpy.fft.

Reference PR

Implementation is available in PR #20952.

Triage

• needs-triage
• frontend:onnx