Device::poll with Maintain::Wait returns early and with an incorrect value when a 5-second timeout is reached

[KSchmidAmilar]

We're using wgpu to do some computing tasks where individual compute passes will take more than CLEANUP_WAIT_MS. In the case of a long running compute pass, waiting on the submission index with a Maintain::Wait or Maintain::WaitForSubmissionIndex will return after the timeout is exceeded, even if the task is not complete. It will also return true indicating the queue is empty if there's nothing else in the queue.

This timeout should be configurable. If that's hard, in the meantime, if this timeout is reached, device::poll call should return false to indicate there is still an item in the queue. Regardless, this behavior should be documented.

[cwfitzgerald]

Seems like there are two issues here:

• We should gracefully and correctly handle timeout failures
• We should expose an optional timeout as part of the Wait/WaitForSubmissionIndex enum

If you're up for a PR, neither should be terribly difficult.