Sparse interferometer inversion: support linear func lists and multiple mappers

[Jammy2211]

Summary

InversionInterferometerSparse now supports an AbstractLinearObjFuncList (e.g. linear light profiles) fitted simultaneously with one or more Mappers, matching what InversionImagingSparse already does. Previously curvature_matrix returned only the first mapper's diagonal block; since the factory routes any list containing a mapper to the sparse class whenever dataset.sparse_operator is set, a mixed list silently dropped every linear-function and cross-mapper term from F.

Every block of F is now built with the same operator W~ = Re(Fᴴ W F) applied by InterferometerSparseOperator on the extent grid: mapper–mapper diagonal (A_iᵀ W~ A_i, unchanged), mapper–mapper off-diagonal (A_iᵀ W~ A_j), mapper–function (A_iᵀ W~ B_k) and function–function (B_kᵀ W~ B_l). Because W~ already contains the inverse-variance weighting, linear-function columns are passed un-weighted (unlike the imaging operator, which is split as Hᵀ N⁻¹ H). The data_vector = Lᵀ d~ expression already covered linear functions exactly and is unchanged.

Also applies no_regularization_add_to_curvature_diag_value as the dense interferometer path does, and fixes __init__ overwriting the parent's settings or Settings() with None when no settings were passed.

Closes #499 (reported by @HRSAstro).

API Changes

Purely additive. Four new methods on InterferometerSparseOperator and func-list / multi-mapper support in InversionInterferometerSparse.curvature_matrix. The single-mapper sparse path is bit-identical to before. One behaviour change: an unregularized mapper in a sparse interferometer inversion now gets the no_regularization_add_to_curvature_diag_value diagonal add (0.001), as the dense path always did.
See full details below.

Test Plan

• pytest test_autoarray/inversion — 377 passed (368 before; 9 new)
• pytest test_autoarray — 1246 passed
• Parity vs InversionInterferometerMapping (7x7 mask, Delaunay, TransformerDFT): func+mapper, mapper+func, two mappers, func+two mappers — curvature_matrix / data_vector max rel err ~6e-16 with unit noise; with a non-uniform noise map the residual (1.2e-9) equals the pre-existing single-mapper baseline, i.e. it is the operator's numerics, not the new blocks.
• JAX (xp=jnp) path bit-identical to NumPy on all multi-object cases
• Downstream: PyAutoGalaxy -k "interferometer or inversion" 68 passed; PyAutoLens 65 passed

Full API Changes (for automation & release notes)

Added

• InterferometerSparseOperator.curvature_matrix_off_diag_from(rows0, cols0, vals0, rows1, cols1, vals1, *, S0, S1) — mapper–mapper off-diagonal block A0ᵀ W~ A1
• InterferometerSparseOperator.curvature_matrix_off_diag_func_list_from(curvature_weights, extent_index_for_masked_pixel, rows, cols, vals, *, S) — mapper–function block Aᵀ W~ B
• InterferometerSparseOperator.curvature_matrix_func_list_from(curvature_weights_0, curvature_weights_1, extent_index_for_masked_pixel) — function–function block B0ᵀ W~ B1
• InterferometerSparseOperator.operated_matrix_slim_from(matrix_slim, extent_index_for_masked_pixel) — scatter slim → extent, apply W~, gather back
• InversionInterferometerSparse._sparse_triplets_curvature_from, _curvature_matrix_mapper_diag, _curvature_matrix_off_diag_from, _curvature_matrix_multi_mapper, _curvature_matrix_func_list_and_mapper (private)

Changed Behaviour

• InversionInterferometerSparse.curvature_matrix — now correct for any mix of AbstractLinearObjFuncList and Mapper objects; previously returned only the first mapper's block. Applies settings.no_regularization_add_to_curvature_diag_value to unregularized linear objects (matches InversionInterferometerMapping).
• InversionInterferometerSparse.__init__ — no longer overwrites self.settings with None when settings is not passed.

Still unsupported

• A linear object with an operated_mapping_matrix_override (visibility-space) still raises InversionException on the sparse path, as before.

Generated by the PyAutoLabs agent workflow.

🤖 Generated with Claude Code

https://claude.ai/code/session_012JM45sA4YGEUw6KYW8Pm96

[Jammy2211]

Workspace PRs (library-first gate — merge after this PR): PyAutoLabs/autolens_workspace_test#283, PyAutoLabs/autogalaxy_workspace_test#115