fix(p_loraks.py): remove incorrect S-matrix operator_shape doubling i…

[wzlk655]

Pull Request Description

Problem

When using P-LORAKS with the S-matrix operator type and multi-coil data, reconstruction failed with the error:

RuntimeError: shape '[125000, 98]' is invalid for input of size 3062500

This occurred in s_operator() at the line indices_rev = indices.view(matrix_shape).flip(dims=(0,)).

Root Cause Analysis

Complete data-flow trace through the S-matrix pipeline:

Step	Object	Elements	Shape
1. get_linear_indices() returns	indices	3,062,500	(62500, 49) = (ns, nb)
2. _initialize_matrix_indices() incorrectly doubles	matrix_operator_shape	—	(125000, 98) = (2×ns, 2×nb)
3. s_operator() calls indices.view(matrix_shape)	indices	3,062,500	→ needs 12,250,000 ❌

The s_operator function internally handles S-matrix doubling via quadrant concatenation:

s_u = concatenate([(s_p - s_m).real, -(s_p - s_m).imag], dim=1)
s_d = concatenate([(s_p + s_m).imag, (s_p + s_m).real], dim=1)
s  = concatenate([s_u, s_d], dim=-1)
# Final: [..., 2×nb, 2×ns] → correct S-matrix size

No external doubling of matrix_operator_shape is needed — the doubling inside s_operator already produces the correct (2×nb, 2×ns) dimensions.

Fix

File: p_loraks.py

Removed the 2 lines that incorrectly doubled matrix_operator_shape for S-matrix:

 def _initialize_matrix_indices(self):
     self.indices, self.matrix_operator_shape = get_linear_indices(
         self.k_space_shape[1:],
         self.patch_shape,
         self.sample_directions
     )
-    if self.operator_type == OperatorType.S:
-        self.matrix_operator_shape = tuple(2 * d for d in self.matrix_operator_shape)