Fix Bayesian detector device guard

[sylvesterkaczmarek]

Fixes #26.

Summary

Correct the Bayesian detector device guard so CPU execution is rejected as intended while GPU/TPU execution is allowed.

The previous condition was inverted: it raised when the device was CUDA or TPU, even though the accompanying error states that CPU training is unstable and recommends GPU or TPU. This changes the predicate to raise only when the device is neither CUDA nor TPU.

Testing

• full repository test suite: 18 passed
• existing PyTorch batching warnings only; no test failures

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[sylvesterkaczmarek]

Could a maintainer review the Bayesian-detector device guard and arrange fresh approved CI and security runs? Both previous runs ended without creating any jobs after 30 days, which looks like expired approval.

[sylvesterkaczmarek]

Following up on #46, #53, #50. Could a maintainer take a look when available? Please let me know if any need changes or a different scope.