sampler and span processor prototype implementations for consistent sampling

[oertl]

Description:

This PR adds various Sampler and SpanProcessor prototype implementations for consistent sampling as defined by https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/tracestate-probability-sampling.md and open-telemetry/opentelemetry-specification#2047:

• ConsistentSampler: abstract base class of all consistent sampler implementations below
• ConsistentAlwaysOffSampler: see https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/tracestate-probability-sampling.md#always-off-sampler
• ConsistentAlwaysOnSampler: see https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/tracestate-probability-sampling.md#always-on-consistent-probability-sampler
• ConsistentComposedAndSampler: allows combining two consistent samplers and samples when both samplers would sample
• ConsistentComposedOrSampler: allows combining two consistent sampler and samples when at least one of both samplers would sample, see https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/tracestate-probability-sampling.md#requirement-combine-multiple-consistent-probability-samplers-using-the-minimum-p-value
• ConsistentParentBasedSampler: see https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/tracestate-probability-sampling.md#parentconsistentprobabilitybased-sampler
• ConsistentProbabilityBasedSampler: see https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/tracestate-probability-sampling.md#consistentprobabilitybased-sampler
• ConsistentRateLimitingSampler: a rate limiting sampler based on exponential smoothing that dynamically adjusts the sampling probability based on the estimated rate of spans occurring to satisfy a given rate of sampled spans
• ConsistentReservoirSamplingBatchSpanProcessor: consistent reservoir sampling of spans, see documentation in the source code

EDIT: ConsistentReservoirSamplingBatchSpanProcessor was removed to make this PR smaller as requested by @anuraaga. It can now be found in https://github.com/dynatrace-oss-contrib/opentelemetry-java-contrib/tree/consistent-reservoir-sampling.

[anuraaga]

Did a very quick skim - 3700 lines is too huge though, let's start with a PR that only adds ConsistentProbabilityBasedSampler please

[oertl]

@trask, thank you very much for your valuable suggestions! I hope I have not overlooked anything and have addressed all your points.

[trask]

@open-telemetry/java-contrib-approvers let me know if anyone is planning to review this further, otherwise I will merge before making the upcoming release, thx!

[anuraaga]

It looks like RandomGenerator can be moved into the samplers package and made package private? We can have package-private factories accepting it for use in unit tests only in this repo, but let's keep the public API using just a default threadlocalrandom based generator without user configuration.

For OtelTraceState it also looks like it can be made package private. If you expect other packages needing the utility methods, then it's also ok to move into io.opentelemetry.contrib.samplers.internal but suspect we don't need to do more than package private at least at this time.

Note that we cannot have packages present in multiple artifacts in our codebase. So currently the generic sounding packages io.opentelemetry.contrib.state and io.opentelemetry.contrib.util could cause problems in the future. The above internalizing suggestions will hopefully fix that naturally.

[anuraaga]

Thanks!

[oertl]

@anuraaga, thanks for your suggestions. I addressed them in da42018

[trask]

thanks @oertl 🎉