Honor omitted weights and masks in single-mode CP reconstruction - #703

Fixes #702.

The one-mode matrix-factor branch of cp_to_tensor returns before the existing implicit-weight fallback and mask application. Omitted weights therefore raise TypeError, and explicit weights with a mask silently produce the unmasked vector.

Move the unit-weight fallback before this branch and apply the supplied mask to its result. Four compact regression/control cases use literal expected vectors and exercise both tuple and CPTensor inputs. They cover a single factor matrix with two component columns; the rank-one vector-factor path from #653 remains separate.

Validation on Windows, CPython 3.12.14, NumPy 2.3.5 and SciPy 1.16.3:

• Full NumPy suite, including available sparse tests: 981 passed, 12 skipped, 1 xpassed, 35 warnings. The unmodified base has 977 passed with the same skip/XPASS/warning counts. The four new cases are included in these totals.
• Unmodified NumPy and PyTorch focused regressions: 3 failed, 1 passed in each; the fix passes all four.
• CP tensor module: 72 passed with NumPy/einsum, PyTorch/core and PyTorch/einsum. These overlapping runs are not additional full-suite cases.
• Black 26.10.0 checks all 156 source/test files successfully.

PyTorch 2.8.0 was tested on CPU. Existing environments and dependencies were reused without changes. The other backend, Python-version and operating-system matrices, documentation and package builds were not run locally.