Better tracking of RTT noLoad

• For the 'gradient' limiter allow for probing for an up to date noload RTT
• Less aggressive min rtt threshold of (1 us) now that we can probe for better RTT
• Make the sample window size relative to the last min RTT and not the absolute min RTT to get a more meaningful sample size
• Track metrics as distributions and not gauges for better visibility

Capture data from sample window in a single immutable class

Merge pull request #38 from elandau/feature/api

Capture data from sample window in a single immutable class

Fix Vegas smoothing

Merge pull request #36 from elandau/bugfix/vegas_smoothing

Fix Vegas smoothing and set default to 1.0

Fix partitionByPathInfo

#32

v0.0.19: Merge pull request #27 from Netflix/feature/rtt_tolerance

• Guages for min_window_rtt and queue_size
• rtt change tolerance configuration

Fix min rtt sample units

Merge pull request #25 from Netflix/feature/min_threshold

Min rtt threshold should be in nanos

Ignore unusually low RTT samples

Merge pull request #24 from Netflix/feature/min_threshold

Eliminated unusually low values that skew the limit estimation

Speed up gradient convergence to the limit

Merge pull request #23 from Netflix/feature/vegas_pre_extend_limit

Gradient algo improvements

Speed up gradient convergence to the limit

Merge pull request #23 from Netflix/feature/vegas_pre_extend_limit

Gradient algo improvements

Square root function optimization

Merge pull request #22 from Netflix/feature/fast_sqrt

Optimized function for calculating square roots