[fix] Replace pytest.skip with requirement_not_met in conjugateGradientMultiBlockCG example (#1530)

[edenfunf]

What

Replace all pytest.skip() calls with requirement_not_met() in
cuda_bindings/examples/4_CUDA_Libraries/conjugateGradientMultiBlockCG_test.py,
and remove the import pytest that was placed inside main().

Before:

def main():
    tol = 1e-5

    import pytest

    # WAIVE: Due to bug in NVRTC
    return

    if platform.system() == "Darwin":
        pytest.skip("conjugateGradientMultiBlockCG is not supported on Mac OSX")
    ...
    if not device_prop.managedMemory:
        pytest.skip("Unified Memory not supported on this device")

After:

def main():
    tol = 1e-5

    # WAIVE: Due to bug in NVRTC
    requirement_not_met("conjugateGradientMultiBlockCG is currently waived due to a known NVRTC issue")

    if platform.system() == "Darwin":
        requirement_not_met("conjugateGradientMultiBlockCG is not supported on Mac OSX")
    ...
    if not device_prop.managedMemory:
        requirement_not_met("Unified Memory not supported on this device")

Why

Issue #1530 tracks that pytest must not be imported in standalone examples.
After auditing all examples in cuda_bindings/examples/, this is the only
file that still imports pytest. Every other example already uses
requirement_not_met() from cuda.bindings._example_helpers to signal
unsupported configurations and exit cleanly via sys.exit.

Using pytest.skip() in a standalone script is incorrect — it raises
pytest.skip.Exception which is only meaningful when running under a pytest
test session. Outside of pytest, the exception propagates uncaught.

How

• Added requirement_not_met to the _example_helpers import at the top of
  the file (it is already exported from cuda.bindings._example_helpers.__all__).
• Removed the import pytest statement that was placed inside main().
• Replaced the silent return waive (for the known NVRTC issue) with a
  proper requirement_not_met() call that prints a message to stderr and
  exits with a non-zero code (or zero if CUDA_BINDINGS_SKIP_EXAMPLE=0).
• Replaced all remaining pytest.skip(...) calls with requirement_not_met(...),
  matching the pattern used in every other example file.

Test

• ruff check and ruff format --check pass on the modified file.
• pre-commit run --files cuda_bindings/examples/4_CUDA_Libraries/conjugateGradientMultiBlockCG_test.py
  passes all hooks (ruff, SPDX headers, end-of-file, trailing whitespace, etc.).
• The file has no remaining references to pytest (verified with grep).
• Behavior is unchanged: the example exits early due to the NVRTC waive,
  just now through the standard requirement_not_met() path instead of
  a silent return.

Fixes #1530
Relates to #1839

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[edenfunf]

Hi, could someone help review and trigger CI for this PR? Thanks!

[rwgk]

Sorry we cannot take PRs against cuda-bindings, which has SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE (only cuda-core and cuda-pathfinder, which have SPDX-License-Identifier: Apache-2.0). Unfortunately I don't have a good idea for making that more obvious; I'll ask around.

[edenfunf]

Sorry we cannot take PRs against cuda-bindings, which has SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE (only cuda-core and cuda-pathfinder, which have SPDX-License-Identifier: Apache-2.0). Unfortunately I don't have a good idea for making that more obvious; I'll ask around.

Thanks for the clarification!

Got it — I didn’t realize that cuda-bindings is not open for external contributions due to the license. Appreciate the explanation.

I’ll focus on cuda-core / cuda-pathfinder or other open components instead.

Thanks again!