CudaClaw : Add timers to time CUDA operations

[donnaaboise]

We should add performance timers to time CUDA memory allocation and patch updates.

See fclaw_timer.h and fclaw_timer.c for templates on how to add timers.

I suggest timers : FCLAW2D_TIMER_CUDA_ALLOCATION, FCLAW2D_TIMER_CUDA_KERNEL1, FCLAW2D_TIMER_CUDA_KERNEL2 and FCLAW2D_TIMER_CUDA_KERNEL3.

• These timers will be accounting for time already accounted for in ADVANCE. So, they should be added in the timer enum after 'UNACCOUNTED, where all non-exclusive timers go. Maybe between FCLAW2D_TIMER_COMM and FCLAW2D_TIMER_EXTRA1 ?

[donnaaboise]

I've added timers, but they still need to be started /stopped from the cudaclaw5 code. This will require updating the arguments lists to step2_wrap and step2.

[donnaaboise]

@MelodyShih added cudaStreams, which are needed for asynchronous calls.