Fix silently ignored CUDA errors in nvcuda backend

[stijnh]

Kernels that fail to launch are silently ignored by the nvcuda backend instead of raising an exception.

cuda-python returns a tuple from every call, even for calls that only return a status, e.g. cuLaunchKernel returns a tuple of length 1 (CUresult,). cuda_error_check assumes that its input is the error code, not a tuple. As a result, kernels that failed to launch were just ignored.

This PR changes cuda_error_check to unwrap tuples before checking the status. It is also raises a TypeError when given anything that is not a CUDA driver, runtime or NVRTC status.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud