Optional OpenTelemetry bridge for Continuum telemetry events.

Continuum does not depend on OpenTelemetry. Applications that already use OpenTelemetry can add the :opentelemetry and :opentelemetry_api packages, configure their exporter, and call setup/1 from their application startup.

{:ok, _handler_id} = Continuum.OpenTelemetry.setup()

The bridge turns existing [:continuum, ...] telemetry events into short spans:

• a continuum.run_attempt span for each engine attempt, closed when the run suspends, completes, fails, is cancelled, or loses its lease
• a continuum.activity_attempt span for each activity attempt, closed when the activity completes, fails, or is retried
• a continuum.compensation_attempt span for each compensation attempt, closed when the compensation completes or fails

Child-workflow and continue_as_new events are recorded as breadcrumbs on the parent/originating run-attempt span; a child's own execution is already captured by its own continuum.run_attempt spans, correlated by run id.

Run-attempt spans always include continuum.run_id. When the run metadata contains a persisted W3C traceparent, the span also gets a link to that original trace context so resumed attempts can be correlated across process and VM restarts.

In durable Postgres mode, activity attempts run in worker processes after the engine attempt has suspended. Those activity spans are usually sibling roots correlated by continuum.run_id, not children of the run-attempt span.