[Grouped Matmul] Fix PyTorch memory leak when tensors are not contiguous

[Jokeren]

For example, if other[i] was transposed, a new tensor is created in the loop of for (size_t i = 0; i < num_matrices; ++i) {.
The data_ptr of the tensor newly created will be used by a kernel following, but this tensor itself may get released before the kernel launch.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (160d5c3) 86.47% compared to head (84b4324) 86.47%.

❗ Current head 84b4324 differs from pull request most recent head 11be60f. Consider uploading reports for the commit 11be60f to get more accurate results

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #290   +/-   ##
=======================================
  Coverage   86.47%   86.47%           
=======================================
  Files          35       35           
  Lines        1213     1213           
=======================================
  Hits         1049     1049           
  Misses        164      164           

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

[Jokeren]

@rusty1s

Partial credit goes to my student @Karthikg99 who reported the problem.

[rusty1s]

Thank you :)