Add head sampling section + elaborate tail sampling

[cartermp]

Fixes #2457

This still leaves our sampling docs in a very, very incomplete state. But in the name of incremental progress, here's something.

Preview: https://deploy-preview-2463--opentelemetry.netlify.app/docs/concepts/sampling/

[kentquirk]

This is definitely an improvement! A few nits, but otherwise I like it.

[chalin]

I'd avoid capitalizing "head sampling" and "tail sampling" (unless the context asks for it like a heading or the start of a sentence).

[svrnm]

can we link to the spec of Consistent Probability Sampling here?

[cartermp]

Doneski!

[svrnm]

non-blocking comment, great addition:)

[cartermp]

/easycla

[cartermp]

Hmmm, after I hit the button I wondered if the head sampling bit should not use the Consistent Probability sampler terminology, since technically SDKs don't implement this yet.

...but, like, the concept is still the same from the perspective of an end-user. Deterministically taking a fraction of traces is what both do, but the CPS propagates metadata that tells you information about the total population that the sample comes from. This makes the CPS superior because it lets a telemetry backend re-weight counts so that any aggregation you derive from tracing data (like total # of requests) is actually correct w.r.t. the total population compared to the sample.

...anyways, I'm really not too worried at the moment, as it's easy enough to change. Frankly this is the direction all the head samplers should be heading anyways.

[jmacd]

LGTM