[improvement] Don't impose DO_NOT_SAMPLE on downstream services

[iamdanfox]

Before this PR

If a service decides to start sampling a trace, it's important that all downstream service do the same to ensure we get a nice complete set of data in trace-log-viewer.

HOWEVER, if the TraceSampler decides not to sample a trace, it currently imposes this DO_NOT_SAMPLE decision on all downstream services too by setting the X-B3-Sampled header to 0

This means that theoretically if a single service early in a call graph has a low sampling rate, then it may unnecessarily reduce tracing information that all downstream services, even if they had plenty of capacity to record traces.

After this PR

If a service decides not to observe a span, it no longer forces all the downstream services to also drop the span, and instead allows downstream samplers to decide.

Longer-term, I'd also quite like to make our trace sampling to be a target logging rate (e.g. up to 1000 lines/second) rather than just a constant percentage of incoming requests. This should mean that infrequently accessed services could just sample every network request, thereby hopefully making life easier for developers.

Possible downsides

• Rolling out this change will probably result in a higher proportion of traces getting sampled, as suddenly a request that passes through will have a chance to be sampled in every service, whereas previously the first failure would effectively blacklist that trace forever.
• I also don't really understand why you would want to mark a trace as explicitly DO_NOT_SAMPLE, so maybe I'm missing something here.

Related to #32

[ferozco]

As you noted this change will lead to a dramatic increase in observed traces. I worry that it may have negative impact on the signals processing pipeline since we are now sampling a request with probability p = sum_i=0^n .98^i * .02 where n is the number of hops. For a request with 5 hops that translates to a sampling rate of 11.4%

[carterkozak]

I think this makes it more difficult and dangerous to configure the trace rate because of the consequences for all downstream services.

[iamdanfox]

Ok fair enough let's park this one