Fixes for blocking limiter and perf improvements

[marcoferrer]

I was writing a benchmark for the blocking limiter when I realized that it was susceptible to deadlocks.

This PR introduces the following changes

• Mild refactor of BlockingLimiter and DeadlineLimiter
  • This reduces the number of goroutines needed to block down to 1
  • The blocking code now creates a timer and explicitly releases it when no longer needed so that it can be reclaimed without needing to elapse
  • Added a benchmark for blocking limiter
• Reduce overhead of common metrics sampler in scenarios where the metrics registry is not supplied or noop.
• Refactor SettableLimiter to reduce the number of mutex checks

[platinummonkey]

thanks!