add a producer-consumer benchmark

[drossetti]

strawman design:

• allocate device memory buffer B
• launch CUDA kernel:
  • polling on B[0]
  • writing a zero-copy flag
• CPU:
  • wait for the kernel to really be polling
  • read tsc in t_start
  • write B[0]
  • wait for flag
  • read tsc in t_end
  • d_t = t_end - t_start should be lower than 1-2 msecs
• repeat until result is stable

[drossetti]

@pakmarkthub regarding #101, note that this issue is really about a different kind of latency test, i.e. not a copy latency. it is meant to measure the latency of writing to device memory through a GDRcopy mapping.