refactor: one kernels module instead of five copies

[Hendrik-code]

Stacked on #60. First of the three deduplication PRs; no new behaviour, but see the caveat below — it is not bit-for-bit everywhere.

What was duplicated

Four independent 3-D Gaussian implementations, three bias fields and two copies of the Laplace/Scharr constant tables, across five modules:

	gpu/contrast.py	gpu/fromSeg.py	gpu/domain_transfer.py	synthseg/functional.py	cpu/contrast.py
Gaussian kernel	✓	✓	✓	✓
Bias field	✓ (polynomial)		✓	✓
Laplace / Scharr	✓				✓

Each carried some version of a comment saying it was kept local so its module would stay self-contained. The synthseg and domain_transfer bias fields were line-for-line identical.

That divergence is the point. It is what let two of the copies go wrong unnoticed — the uncentred Gaussian (#58) and the malformed 2-D Scharr x-kernel (#58) — and why each had to be fixed on its own rather than in one place. smauglab/transforms/kernels.py is that one place now.

Not bit-for-bit, deliberately

Two blur call sites change slightly:

• Radius is ceil(3*sigma). domain_transfer and fromSeg used round(3*sigma), which is never wider, so their kernels may now be one tap larger.
• Padding is reflect everywhere. domain_transfer used replicate; fromSeg relied on conv3d's implicit zero padding, which pulls the volume border towards 0. That is the one difference here that was wrong rather than merely different.

Everything else is the same arithmetic as the copy it replaces.

Interfaces are unchanged

The local wrappers stay where their callers expect them — _gaussian_blur_3d in fromSeg.py still clamps to [0, 1], _random_bias_field3d in domain_transfer.py still returns a bare [D, H, W] volume rather than [B, C, D, H, W]. This is a change of implementation, not of interface.

Testing

New unit_tests/test_kernels.py (12 tests) covers the merged behaviour and, more usefully, the equivalences that make this one implementation rather than a sixth copy:

• test_synthseg_and_domain_transfer_now_share_one_implementation — same seed, same field
• test_the_kernels_are_shared_by_both_backends — the CPU transform hands out the shared tables
• test_a_constant_volume_survives_the_blur — the reflect-padding property that the zero-padded copy failed

test_kernel_correctness.py from #58 follows the Gaussian to its new home; its 1-D assertions move onto the sigma-derived gaussian_kernel1d, since the fixed-size helper is now internal to gaussian_kernel3d.

Net: −209 / +56 across the five modules, plus the 265-line shared module.

Full suite: 82 passed, 169 subtests. ruff and mypy clean.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.