Add norm parameter to FFT transforms (backward/ortho/forward)

[Aristide021]

Proposed changes

Adds norm support ("backward", "ortho", "forward") to FFT transforms for NumPy parity.

Matches NumPy numpy.fft norm conventions and addresses the normalization scaling gap noted in the mlx/primitives.cpp FFT VJP TODO.

• Adds FFTNorm to core FFT API and threads it through fft/ifft/rfft/irfft, fft2/ifft2/rfft2/irfft2, fftn/ifftn/rfftn/irfftn
• Implements normalization scaling in mlx/fft.cpp:
  • "backward": existing behavior (default)
  • "ortho": 1/sqrt(N) forward, sqrt(N) inverse adjustment
  • "forward": 1/N forward, N inverse adjustment
• Adds Python norm="backward" parameter for all 12 FFT APIs with input validation
• Adds NumPy parity tests for all norm modes and updates FFT docs

Default behavior is unchanged (norm="backward").

Checklist

• I have read the CONTRIBUTING document
• I have run pre-commit run --all-files to format my code / installed pre-commit prior to committing changes
• I have added tests that prove my fix is effective or that my feature works
• I have updated the necessary documentation (if needed)

[zcbenz]

The StreamOrDevice parameter should always be put last.

[Aristide021]

The StreamOrDevice parameter should always be put last.

Moved StreamOrDevice to last across all FFT APIs and call sites.

[zcbenz]

Thanks for the contributions! This basically looks good to me, just a few more small requests.

[Aristide021]

The test_vmap_masked_scatter failure appears unrelated to this PR,
it tests vmap masked scatter behavior which is not touched by these
changes. May be a pre-existing flaky test. @zcbenz