cMeta AOPS v0.61.4

• A framework's own report in --provenance=loaded: after an observed run the recorder asks torch
  (torch.version.cuda, cuDNN, the CUDA devices it sees, MPS), ONNX Runtime (device, providers) and JAX
  (backend, devices) what they report about themselves - through the run's own interpreter by its full
  path, with the run's environment, in the build folder, one process per framework with a 90 s limit -
  and records it under runtime.frameworks[<dist>].reports (the interpreter and the seconds under
  .probe). A torch that reports another CUDA than its wheel was built for, or no CUDA device in a cuda
  run, is a warning (framework); a probe without an answer is a note. on and strict start no process.
  Tests: test_framework_probes_use_the_runs_python, test_framework_report_checks.