Add the random fixed size exemplar reservoir

[MrAlias]

Part of #559

This PR is split from #4455

This adds a Reservoir implementation that will randomly sample a specified number of measurements as exemplars. An explanation of the algorithm is copied from the included comment within the PR:

The following algorithm is "Algorithm L" from Li, Kim-Hung (4 December 1994). "Reservoir-Sampling Algorithms of Time Complexity O(n(1+log(N/n)))". ACM Transactions on Mathematical Software. 20 (4): 481–493 (https://dl.acm.org/doi/10.1145/198429.198435).

A high-level overview of "Algorithm L":

0. Pre-calculate the random count greater than the storage size when an exemplar will be replaced.
1. Accept all measurements offered until the configured storage size is reached.
2. Loop:
   a) When the pre-calculate count is reached, replace a random existing exemplar with the offered measurement.
   b) Calculate the next random count greater than the existing one which will replace another exemplars

The way a "replacement" count is computed is by looking at n number of independent random numbers each corresponding to an offered measurement. Of these numbers the smallest k (the same size as the storage capacity) of them are kept as a subset. The maximum value in this subset, called w is used to weight another random number generation for the next count that will be considered.

By weighting the next count computation like described, it is able to perform a uniformly-weighted sampling algorithm based on the number of samples the reservoir has seen so far. The sampling will "slow down" as more and more samples are offered so as to reduce a bias towards those offered just prior to the end of the collection.

This algorithm is preferred because of its balance of simplicity and performance. It will compute three random numbers (the bulk of computation time) for each item that becomes part of the reservoir, but it does not spend any time on items that do not. In particular it has an asymptotic runtime of O(k(1 + log(n/k)) where n is the number of measurements offered and k is the reservoir size.

See https://en.wikipedia.org/wiki/Reservoir_sampling for an overview of this and other reservoir sampling algorithms. See https://github.com/MrAlias/reservoir-sampling for a performance comparison of reservoir sampling algorithms.

[codecov]

Codecov Report

Attention: 3 lines in your changes are missing coverage. Please review.

Comparison is base (ce3faf1) 82.3% compared to head (bd024cb) 82.5%.

Additional details and impacted files

@@           Coverage Diff           @@
##            main   #4852     +/-   ##
=======================================
+ Coverage   82.3%   82.5%   +0.1%     
=======================================
  Files        228     230      +2     
  Lines      18569   18754    +185     
=======================================
+ Hits       15300   15480    +180     
- Misses      2981    2985      +4     
- Partials     288     289      +1     

Files	Coverage Δ
sdk/metric/internal/exemplar/storage.go	100.0% <100.0%> (ø)
sdk/metric/internal/exemplar/rand.go	97.7% <97.7%> (ø)

... and 1 file with indirect coverage changes