Document cleanup()

[ptheywood]

Add documentation about using flamegpu::util::cleanup() or pyflamegpu.cleanup() once at the end of the applciation to allow/ensure profiling / cuda memcheck to work correctly, including the dangers of calling the method too early / from the host API.

[Robadob]

More generally, this is the only discussion we currently have of performance. Need to have a think about what might go in a profiling page (as we don't really want to provide a guide for how to use nvidia tools, but will also want to mention NVTX).