* cuda.cccl: Update dependencies to enable running on CUDA 13 driver (#5442)

* Add numba-cuda as a dependency

* Replace use of pynvjitlink patch

* Update pyproject.toml

There's a bug in cuda-bindings 12.9.0 that prevents us from using CUDA 13 driver

* Pathfinder update

* Remove all uses of CUDA_ENABLE_PYNVJITLINK. Remove pynvjitlink dependency. Add numba-cuda lower bound.

* Remove all empty imports sections

* Restore some imports sections

* It's lib, not library

---------

Co-authored-by: Ashwin Srinath <shwina@users.noreply.github.com>

* We need cuda.core?

* We need cuda.core?

* Add begin/end markers to warp merge sort API example

---------

Co-authored-by: Ashwin Srinath <shwina@users.noreply.github.com>

The sync was too late and did not guard loading from merge_partitions, leading to a data race

This is a partial backport of #5414

Fixes NvBug 5320137 and NVIDIA/cccl_private#453

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

Co-authored-by: David Bayer <48736217+davebayer@users.noreply.github.com>

We are getting QA bugs because of those failures, most are MSVC limitations that are popping up again.

There is one fix in the definition of `indirectly_writable` for an MSVC oddity

We were experiencing segfaults with `submdspan` on MSVC

Thanks to some friendly help of the compiler team we have a workaround for those crashes

* Backport pin of llvmlite dependency to branch/3.0x

* Pin llvmlite dependency in pyproject.toml for cuda_coop

Added llvmlite dependency with a note to remove it later.

* [pre-commit.ci] auto code formatting

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

(cherry picked from commit 65c0373)

Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>

…omplex<T>` (#6102) (#6117)

* Use proper qualification in allocate.h (#4796)

Fixes [BUG]: Ambiguous call to __do_deallocate_handle_size when compiling with clang and libc++ #4793

* Keep sink for `__align`

---------

Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>