Do not validate tracestate

[justinfoote]

The opentelemetry-java implementation of W3C TraceContext is currently validating every tracestate entry (See here). An invalid tracestate entry will cause an IllegalArgumentException, via Utils::checkArgument, which will bubble up to HttpTraceContext::extract, here. That exception causes the entire W3C TraceContext payload to be discarded!

The W3C spec allows for tracestate validation, with this text:

The vendor MAY validate the tracestate header. If the tracestate header cannot be parsed the vendor MAY discard the entire header. Invalid tracestate entries MAY also be discarded.

But I don't believe it intends for vendors to discard traceparent and restart the trace.
In addition, there is an open issue (w3c/trace-context#384) to change that to a SHOULD NOT, rather than a MAY.

I propose that this project remove all validation of tracestate entries.

Or, at the very least, I propose that this project drop only invalid tracestate entries, rather than the entire W3C TraceContext payload.

[jkwatson]

See open-telemetry/opentelemetry-specification#478 for possible spec change.

[carlosalberto]

Related to #895

[bogdandrutu]

@jkwatson lazy parsing and validation of the tracestate is a nice addition to our project. I am happy to see a PR around that idea.

[Oberon00]

I think a better design might be to propagate and store the tracestate more independent of trace+spanid on the Context. That would be a spec change though.

[jkwatson]

I'd like to move this to post-GA, if that's ok with folks. Please let me know if you want me to move it back.

[iNikem]

I would like to point out, that there are 2 different use cases here. One is while reading TraceState from the wire, another is during manipulating TraceState before propagation/export. The second use case is now explicitly addressed by the spec:

Every mutating operations MUST validate input parameters