[NDTensorsCUDAExt] Fix bug in slicing Adjoint CuMatrix

[kmp5VT]

Description

CUDA was numerically failing for DMRG when the number of sites was above 8 this is related to a truncation bug found in SVD. This merge should fix this bug and give DMRG equivalent results on GPU and CPU.

[mtfishman]

Thanks! Could you add a regression test in NDTensors? Also maybe a good time to add a test for DMRG using NDTensorsCUDAExt as we discussed.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (3b1fa88) 85.42% compared to head (571aa42) 54.14%.

❗ Current head 571aa42 differs from pull request most recent head 6f448e0. Consider uploading reports for the commit 6f448e0 to get more accurate results

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #1236       +/-   ##
===========================================
- Coverage   85.42%   54.14%   -31.29%     
===========================================
  Files          89       87        -2     
  Lines        8400     7853      -547     
===========================================
- Hits         7176     4252     -2924     
- Misses       1224     3601     +2377     

see 65 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mtfishman]

Looks good, thanks for sticking with all of the comments!