Make main function generic

[AntonReinhard]

I added some extra info:

• reasonable type information to function headers
• @inbounds macro
• some more ::Int32 constraints to make sure everything is actually an int32
• always_inline = true for CUDA kernel calls
  These changes might improve performance, but I haven't benchmarked it properly yet.

I also made the function generic so it can be used with any type T that supports basic arithmetic. To reflect that, I made the tests generic, too, and now test over some Float, Integer, and Complex types.

Checklist

• I am following the contributing guidelines
• Tests are passing
• Lint workflow is passing
• Docs were updated and workflow is passing

[AntonReinhard]

I have made my own plots in this branch: https://github.com/AntonReinhard/InclusiveScans.jl/tree/plotting

[szabo137]

LGTM, many thanks for providing this! I tested the code in a GV100, and it works. Therefore, I use my admin power to merge this.

Later we might configure GPU tests for this, but for now, this should be fine.