Enable support for externally-defined ID generators

[Aneurysm9]

• Moved the SDK's internal.IDGenerator interface to the sdk/trace
  package.
• Added trace.WithIDGenerator() TracerProviderOption.

[codecov]

Codecov Report

Merging #1363 (13a268d) into master (0021ab0) will decrease coverage by 0.1%.
The diff coverage is 86.6%.

@@           Coverage Diff            @@
##           master   #1363     +/-   ##
========================================
- Coverage    77.7%   77.6%   -0.2%     
========================================
  Files         124     123      -1     
  Lines        6126    6131      +5     
========================================
- Hits         4761    4758      -3     
- Misses       1114    1122      +8     
  Partials      251     251             

Impacted Files	Coverage Δ
sdk/trace/provider.go	92.3% <33.3%> (-2.1%)	⬇️
sdk/trace/id_generator.go	100.0% <100.0%> (ø)
sdk/trace/span.go	94.9% <100.0%> (+<0.1%)	⬆️
sdk/trace/tracer.go	100.0% <100.0%> (ø)
sdk/trace/batch_span_processor.go	74.5% <0.0%> (-5.9%)	⬇️

[jmacd]

I made a note in the associated spec PR (open-telemetry/opentelemetry-specification#1006 (comment)) that I'd probably rather see two methods, where one returns a single SpanID (taking a mutex once) and one returns a pair of TraceID and SpanID (taking a mutex once). What do you think?

[krnowak]

I made a note in the associated spec PR (open-telemetry/opentelemetry-specification#1006 (comment)) that I'd probably rather see two methods, where one returns a single SpanID (taking a mutex once) and one returns a pair of TraceID and SpanID (taking a mutex once). What do you think?

Sounds useful as this seems to be how the ID generator is going to be used. Would there be a case where we only wanted to generate a trace ID?

[MrAlias]

I made a note in the associated spec PR (open-telemetry/opentelemetry-specification#1006 (comment)) that I'd probably rather see two methods, where one returns a single SpanID (taking a mutex once) and one returns a pair of TraceID and SpanID (taking a mutex once). What do you think?

Yeah, I think this is indeed going to be how this will be used.

Would there be a case where we only wanted to generate a trace ID?

I feel like that would be the corner case anyway so throwing away that returned value wouldn't be too much of an issue.

[krnowak]

One nit, otherwise looks good.