fix: drop deleted TransformerNUFFTPyNUFFT from jax_profiling dataset setup

[Jammy2211]

Fixes the only executable reference to TransformerNUFFTPyNUFFT, deleted by PyAutoLabs/PyAutoArray#475, left anywhere in the workspace repos.

The bug

jax_profiling/dataset_setup/interferometer.py:140 named the deleted class. The transformer dict at :137 is built eagerly inside simulate() (:106), before the key lookup — so every instrument raised AttributeError, not just the alma_high_res config that selected the pynufft backend. Reproduced with simulate('sma'), a DFT dataset. All JAX-profiling dataset setup in this repo was broken.

Scripts Changed

Script	Change
jax_profiling/dataset_setup/interferometer.py	Drop the "nufft_pynufft" dict arm; repoint alma_high_res to "nufft"; rewrite the stale rationale comment
jax_profiling/jit/interferometer/delaunay.py	Correct three stale pynufft/JIT claims (prose only)
jax_profiling/jit/datacube/delaunay.py	Correct one stale pynufft/JIT claim (prose only)

No behaviour change in the jit/ scripts — they still use TransformerDFT.

Why alma_high_res becomes NUFFT, not DFT

Both objections in the original comment were checked on 2026-08-23:

• "DFT would need a dense (n_vis × n_real_space) matrix = ~20GB and OOM on a 15GB laptop" — still true, and it rules DFT out. 5000 vis × 512×512 = 1.31e9, far above the ~1e7 n_vis · n_pix crossover measured during the removal work, where the NUFFT is the only feasible path.
• "nufftax requires Python >= 3.12 (PyAutoGPU venv is 3.10)" — obsolete. PyAutoArray, PyAutoLens and PyAutoNerves all declare requires-python = ">=3.12", and nufftax 0.6.1 needs only >=3.11. A 3.10 venv cannot run current autolens at all.

The JIT claim, corrected

Four sites called TransformerNUFFT "pynufft-based" and "not JIT-friendly". Both wrong. The real subtlety, verified here: transform_mapping_matrix takes a per-call xp argument defaulting to np. With xp=jnp it jits and matches the numpy path to 5e-16 relative; with the default it raises TracerArrayConversionError under jax.jit. That is what the caveat now says.

Validation

• simulate() run for every instrument key — sma, alma, alma_high_res, hannah — all pass. The eager dict is exactly why a single-instrument check would not have proved the fix.
• alma_high_res completes without OOM on the nufftax path.
• Alias-aware AST sweep: 0 executable references to the deleted class remain in this repo.
• Generated datasets were reverted — this PR touches source only.

Notes for the reviewer

• Heart was RED (release validation FAILED, unrelated to this change) at PR-open; the human explicitly authorized push + PR-open. Merge remains gated.
• Pre-existing and not addressed here: the committed datasets under jax_profiling/dataset/ no longer reproduce from the current script — regenerating an untouched config (sma) yields different data plus a differing SMALLDAT header stamp. Unfiled; flagged for separate triage.

Closes #128