Reference resource generation CI jobs are failing

[scotts]

See: https://github.com/meta-pytorch/torchcodec/actions/runs/19951109109/job/57245904008?pr=1100

It happens every time, so I think it's not a transient failure. And no other CI jobs encounter it. We may have missed some test-infra related change we did to other CI jobs.