Do not capture timing if there is no work to time.

[ptheywood]

Cuda event timers are used to collect timing information for the various sections of a simulation (init functions, step tim, step functions, exit functions).

Although very cheap, these timers do have a cost (~12us according to nsys to create 2 events, record 2 events, stop 2 events, and syncrhonise them / get the elapsed time).

In cases where there is nothing to do (i.e. if there are no exit functions) then there is no point in timing the exit functions.

For input and exit, this won't help much, but for per-iteratiion recording of step functions if there are none then this might add up to something measurable eventually.

Super low priority as low cost, but would be a very minor efficiency improvement.

[ptheywood]

Currently Step functions are not timed independently of the step itself.

As of #640, RTC, init and exit timers will be steady clock timers, with CUDAEvent timers being used just for the total simulation timer and per iteration time. I don't believe it's possible to have a CUDASimulation which would not contain atleast one function as a part of a step, so I'm going to close this issue.

It was never going to be a significant change either way.