Configurable wait time for operation channels

[Keren Zhou]

There was a channel wait call after the following line.

hpctoolkit/src/tool/hpcrun/gpu/gpu-operation-multiplexer.c

Line 106 in f1236e3

 gpu_operation_channel_set_process(current_operation_channels_count); 

When using cupti continuous pc sampling, we use the operation channel to coordinate application threads and the background thread to attribute pc samples. If the operation channel is called very frequently without waiting, it could cause up to 4x slowdown on ufront.

Therefore, I propose to add a control knob for the wait time to reduce overhead in certain cases. By default the wait time can be 0.

[Keren Zhou]

Seems like the problem comes from CUPTI pc sampling's bug but not our operation thread.