Fix operator precedence in test_cudart.supportsCudaAPI

[LeSingh1]

def supportsCudaAPI(name):
    return name in dir(cuda) or dir(cudart)

This parses as (name in dir(cuda)) or dir(cudart). dir(cudart) is a non-empty list for
any module, so it is unconditionally truthy — the function returns a truthy value for
every input, including names that exist nowhere:

>>> supportsCudaAPI("this_is_not_a_cuda_api")
['__all__', '__builtins__', ...]        # a list, not False

The left operand is dead as well. cuda is cuda.bindings.driver, and every name passed
in is a cudaXxx runtime symbol:

probed name	driver.pyx	runtime.pyx
cudaGraphGetId	0	1
cudaGreenCtxCreate	0	1
cudaDeviceGetExecutionCtx	0	1
cudaGraphConditionalHandleCreate	0	2

So name in dir(cuda) is always False and the result is always the dir(cudart) list.

Consequence

not supportsCudaAPI(...) is always False, so the API-presence half of all 17
skipif guards that use it never fires — lines 1443, 1501, 1528, 1575, 1617, 1646, 1693,
1712, 1727, 1745, 1765, 1793, 1862, 1885, 1902, 1933, 1954. Only the
driver_version_less_than(...) half of each guard does any work.

Those guards exist because the driver version and the binding version are independent:
cuda-bindings built against an older toolkit will not expose cudaGraphGetId even on a
13.1 driver. Today such a build runs the test and dies with AttributeError instead of
skipping.

Fix

name in dir(cuda) or name in dir(cudart) — complete the second comparison. I kept the
dir(cuda) half rather than narrowing to the runtime module: it is what the author wrote,
it is harmless, and narrowing it would be a second, unrelated decision.

Test

test_supportsCudaAPI pins all three cases — a runtime-only name (cudaMalloc), a
driver-only name (cuInit), and a name that exists in neither. The last two fail on
main: supportsCudaAPI("cuInit") returns a list, not True, and
supportsCudaAPI("this_is_not_a_cuda_api") returns that same truthy list instead of
False. It needs no GPU and no driver — dir() does not load anything.

ruff format --check is clean. ruff check reports one pre-existing I001 on this file
that is byte-identical on upstream/main; I left it alone as out of scope.

[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.