Reset trace context after fork

[delner]

When Ruby processes fork, they copy objects from the parent process into the child process: this includes the trace context objects we've assigned to each Thread. As a result, child processes are copying all the spans from the parent process, which could result in redundant transmission of spans.

Instead, we'd like forked processes to act like distributed traces; they should be linked to the parent process (trace ID/parent ID/sampling, etc) but only build and transmit spans generated from its own process.

To do this, we're leveraging the forking helpers implemented in #1221 to make the Datadog::Context fork aware, then lazily clone the Datadog::Context in the Datadog::DefaultContextProvider whenever a context is accessed and observed to have forked.

[delner]

Note, it appears as though this pull request may address an latent issue where forking mid-trace would cause spans generated in the fork to be lost. e.g.

tracer.trace('parent') do
  fork do
    tracer.trace('fork') do
      # some work
    end
  end
end

Before these changes, the fork would inherit all the spans from the parent process, and when the fork would complete, it would think its "still active" thus skip flushing fork spans when the fork ends. After these changes, the parent spans are absent in the fork, allowing spans generated in the fork to be flushed because the trace is "complete".