CI: fix self-hosted cupy action

[mrava87]

This PR implements a workaround to overcome the issues experienced in the self-hosted GA - e.g., https://github.com/PyLops/pylops/actions/runs/26692969903/job/78672544537

FAILED pytests/test_nonstatconvolve.py::test_StationaryConvolve3D[par0] - numba.cuda.cudadrv.driver.LinkerError: [222] Call to cuLinkAddData results in CUDA_ERROR_UNSUPPORTED_PTX_VERSION
ptxas application ptx input, line 9; fatal   : Unsupported .version 8.8; current version is '8.7'

This happens to all tests that use any operator with numba-cuda jit'ed methods.

The issue is likely caused by the fact that CUDA_HOME points to a version of CUDA that is too new for the drivers of the GPU. Instead of relying on Yuxi's CUDA, the action is not installing a self-contained version of CUDA and changing CUDA_HOME to point to it (as well as adding a sym link required by numba)

[codacy-production]

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

_{NEW Get contextual insights on your PRs based on Codacy's metrics, along with PR and Jira context, without leaving GitHub. Enable AI reviewer
TIP This summary will be updated as you push new changes.}