CUDA algorithms should handle empty inputs gracefully

[stephenswat]

Related to #413, the CUDA code currently does not check whether the inputs passed to it are empty. In some cases (e.g. in the seeding) this causes the host to request a kernel launch with zero blocks, which causes a launch failure and breaks the application. For each kernel, we should aim to determine the appropriate behaviour in case the input size is zero, so that we can handle these edge cases gracefully.

[beomki-yeo]

I think we can simply have an if statement which checks the number of inputs

[stephenswat]

Depending on the algorithm we will need to ensure that sane (even if empty) results are passed onto the follow-up algorithms; may not be trivial in all cases.