Remove pynufft and the legacy TransformerNUFFTPyNUFFT backend

Deletes the legacy pynufft-backed interferometer transformer and drops
pynufft from the `optional` and `dev` extras. The nufftax-backed
`TransformerNUFFT` has been the default backend since the nufftax
migration, and `TransformerDFT` remains as the pure-numpy transformer.

Removed:
- `TransformerNUFFTPyNUFFT`, the `NUFFTPlaceholder` / `NUFFT_cpu`
  module-scope try-import, and `pynufft_exception()`
- the `TransformerNUFFTPyNUFFT` re-exports from `autoarray/__init__.py`
  and `autoarray/type.py`, and its arm of the `Transformer` union
- the three `test__nufft_pynufft__*` tests and the `"pynufft"` arm of the
  nufftax-absent skip filter in `test_autoarray/conftest.py`
- `pynufft` from `[project.optional-dependencies]` `optional` and the
  `pynufft==2022.2.2` pin from `dev`

`nufftax_exception()` no longer points users at the deleted class; it now
names `TransformerDFT` as the fallback where JAX is unavailable, and the
`use_adjoint_scaling` docstrings no longer describe a pynufft-specific
normalisation.

This also resolves the pinned `pynufft==2022.2.2` dev-extra failure under
SciPy >= 1.17 (it calls the removed `scipy.linalg.pinv2`).

Note on import time: the measured saving is ~10 ms, not the ~230 ms the
task assumed. `pynufft`'s 0.19 s cumulative import is ~95% `scipy.sparse`,
which `autoarray/operators/derivative_util.py` imports eagerly for
`csr_matrix` regardless. Deferring that import is the real ~0.10 s win and
is filed separately.

Author: claude

autoarray/__init__.py

@@ -64,7 +64,6 @@
 from .mask.mask_2d import Mask2D
 from .operators.transformer import TransformerDFT
 from .operators.transformer import TransformerNUFFT
-from .operators.transformer import TransformerNUFFTPyNUFFT
 from .operators.over_sampling.decorator import over_sample
 from .operators.contour import Grid2DContour
 from .layout.layout import Layout1D