Initialize default device lazily

[janEbert]

This way, we can import the library without requiring a GPU to be present.

What does this PR do?

Do not query the torch.cuda.current_device() upon importing the library. Instead, query it when we actually instantiate the arguments struct.

What issue(s) does this change relate to?

Not submitted.

Before submitting

• Have you read the contributor guidelines?
• Is this change a documentation change or typo fix? If so, skip the rest of this checklist.
• Was this change discussed/approved in a GitHub issue first? It is much more likely to be merged if so.
• Did you update any related docs and document your change?
• Did you update any related tests and add any new tests related to your change? (see testing)
• Did you run the tests locally to make sure they pass?
• Did you run pre-commit on your change? (see the pre-commit section of prerequisites)

[mihir-db]

LGTM!