Update to clang 20 across RAPIDS

[bdice]

With the upgrade to Python 3.13 (#120), we are observing a conda solve failure that can be minimized to:

conda create -n test --dry-run -c conda-forge libclang==16.* matplotlib python=3.13

This can be seen in the CCCL + RAPIDS CI logs and has also been reproduced locally by @jinsolp.

The root cause is:

• matplotlib requires a bunch of GUI libraries like pyside6 (matplotlib-base does not require these)
• The matplotlib versions that support Python 3.13 are built against newer dependencies that require libclang13 >=18.*
• However, we currently require libclang13 ==16.0.6. We pin to 16.0.6 in all conda environments RAPIDS-wide.

Solutions (we should do both):

• Use matplotlib-base in conda environments instead of matplotlib (this is an important change but is ultimately a short-term fix for this problem -- other problems will arise in the future)
  • Use matplotlib-base in docs/notebooks environments. cuml#6723
• Upgrade to a newer libclang / clang / clang-tools version (20.1.4) across RAPIDS. This also requires us to update our pre-commit hooks for clang-format.
  • Update to clang 20 rapids-cmake#846
  • Update to clang 20 rmm#1918
  • Update to clang 20 kvikio#718
  • Update to clang 20 pynvjitlink#143
  • Update to clang 20 cudf#18818
  • Update to clang 20 ucxx#414
  • Update to clang 20 raft#2665
  • Update to clang 20 cugraph#5073
  • Update to clang 20 cugraph-gnn#205
  • https://github.com/rapidsai/cumlprims_mg/pull/248
  • Update to clang 20 cuml#6724
  • Update to clang 20 cuvs#898
  • https://github.com/rapidsai/cuopt/pull/2491