[flyingdoubleG]

Summary

This PR fixes two crashes in inv_quad_logdet.

1. Vector right-hand side through CholLinearOperator

CholLinearOperator.inv_quad_logdet explicitly accepts a 1-D inv_quad_rhs for a non-batch operator, as the default implementation does. It then calls inv_quad, which solves with the vector and sums over dim=-2. For a 1-D solve that dimension does not exist:

IndexError: Dimension out of range (expected to be in range of [-1, 0], but got -2)

Operators below max_cholesky_size compute inv_quad_logdet through a CholLinearOperator. So this also breaks the common call

DenseLinearOperator(A).inv_quad_logdet(vec, logdet=True)   # IndexError on main

while DenseLinearOperator(A).inv_quad(vec) (default path) works and returns a scalar.

Fix. In inv_quad, treat a 1-D right-hand side of a non-batch operator as one column. The output shapes then match the default implementation: a scalar with reduce_inv_quad=True, and (1,) otherwise.

2. CatLinearOperator.logdet()

CatLinearOperator.inv_quad_logdet moves every returned term to the output device, including the None returned for a term that was not requested. So CatLinearOperator.logdet() and inv_quad_logdet(logdet=True) without a right-hand side raise:

AttributeError: 'NoneType' object has no attribute 'to'

Fix. Leave None terms unchanged.

Tests

• TestCholLinearOperator.test_inv_quad_vector: covers inv_quad (reduced and not reduced) and inv_quad_logdet with a vector, plus the DenseLinearOperator path.
• TestCatLinearOperator.test_logdet_without_inv_quad.

Both fail on main and pass with this change. The full suite gives 5049 passed and 1 failed. The failure is test_psd_safe_cholesky_psd, which also fails on main without these changes. I ran ufmt and flake8 on the changed files.

Related, but not covered by this PR: #139 and #147 (vector right-hand sides for DiagLinearOperator batches and low-rank operators), and #144/#145 (CholLinearOperator with upper factors, and slicing).