pynvjitlink: stop development and packaging publishing

[jameslamb]

Description

RAPIDS projects are dropping their pynvjitlink dependency in RAPIDS 25.10.

See rapidsai/pynvjitlink#130 for some important background.

The project should be removed from RAPIDS builds and docs, downstream projects should be notified / migrated, and the pynvjitlink repo should be archived. This tracks that work.

Benefits of this work

• nudges more projects towards cuda-python / numba-cuda
• reduces the effort required for RAPIDS releases
• reduces the need to update pynvjitlink for CUDA 13

Acceptance Criteria

• nothing in RAPIDS depends on pynvjitlink
• [RFC] Archive this project in favor of cuda.core.(experimental).Linker pynvjitlink#130 is archived
• projects depending on pynvjitlink have been migrated or at least notified

Approach

• migrate dependencies away from pynvjitlink ('NVIDIA' GitHub search | 'rapidsai' GitHub search)
  • cudf (Update to numba-cuda>=0.18.0,<0.19.0 NVIDIA/cudf#19604)
  • numba-cuda (Remove remaining references/dependencies on pynvjitlink NVIDIA/numba-cuda#357)
  • numbast (Completely Remove Pynvjitlink Requirements NVIDIA/numbast#159)
  • nvmath-python (@ZzEeKkAa confirmed offline that pynvjitlink will be removed in the next release of nvmath-python, so doesn't need a new release of it)
• update numba-cuda to at least 0.18 across RAPIDS
  • cudf (Update to numba-cuda>=0.18.0,<0.19.0 NVIDIA/cudf#19604)
  • dask-cuda (Update to numba-cuda >=0.18.0,<0.19.0a0 dask-cuda#1533)
  • rmm (Update to numba-cuda >=0.18.0,<0.19.0a0 rmm#2019)
  • ucxx (Update to numba-cuda >=0.18.0,<0.19.0a0 ucxx#486
• stop encouraging the use of pynvjitlink and numba_config.CUDA_ENABLE_PYNVJITLINK in docs (or at least open issues encouraging different patterns)
  • NVIDIA/accelerated-computing-hub (Numba / pynvjitlink updates NVIDIA/accelerated-computing-hub#82)
  • NVIDIA/cccl (remove pynvjitlink references in examples NVIDIA/nvbench#269)
  • NVIDIA/nvbench CCCL examples (remove pynvjitlink references from examples NVIDIA/cccl#5826)
• put up an RSN announcing the plan to stop developing pynvjitlink (RSN 52: dropping pynvjitlink docs#672)
• remove from RAPIDS build infrastructure
  • release-scripts (https://github.com/rapidsai/release-scripts/pull/37)
  • main_branch_repos list in rapids-reviser (https://github.com/rapidsai/rapids-reviser/pull/89)
  • rapids-metadata (drop pynvjitlink from RAPIDS 25.10 rapids-metadata#61)
  • integration (test-conda-nightly-env: remove cubinlinker and pynvjitlink from configuration integration#793)
  • build-planning docs (update cubinlinker, ptxcompiler, pynvjitlink references in release docs #211)
• mark the RSN as completed (mark CUDA 12.9 and pynvjitlink RSNs complete docs#675)
• archive the repo
  • pin an informative issue explaining the archiving ([RFC] Archive this project in favor of cuda.core.(experimental).Linker pynvjitlink#130)
  • put a note in the README about the archiving (add note about the repo being archived pynvjitlink#155)
  • archive https://github.com/rapidsai/pynvjitlink

Notes

See rapidsai/pynvjitlink#130 (comment) for details to include in RSN(s), archival notes, warning messages, etc.

[jameslamb]

I just archived https://github.com/rapidsai/pynvjitlink