Introduce a configurable sample rate

[ethangunderson]

Hello 👋,

The projects where we use Spandex operate at a fairly large scale. A byproduct of this is that it becomes cost prohibitive to send all traces to Datadog. To solve this, we implemented our own Spandex sender that implements a sampling rate to determine trace priority before sending the trace to the Datadog agent.

This works pretty well, but falls apart when we use continue_trace to setup distributed tracing between different processes. Since priority is determined at the time of sending the trace, each trace priority is independently determined instead of the decision being made once and passed via trace context.

To solve for this I've added an optional configuration for a sampling rate. This effectively enables head based sampling as described in the Datadog docs.