Major Description
Noticed this when looking into some tracing things with Billie: it doesn't appear that we have recovery instrumented with tracing.
It would be nice to know what the long pole in the tent is for recovery since it typically represents a period of unavailability of some data for users. We should be aware of why it takes as long as it does and try to reduce it as much as possible.
Because spans are delivered via ZK, I think it will be ok if we're performing recovery on a WAL which contains updates for the trace table. As long as the serialization to the trace table doesn't cause problems (it should just create back-pressure in the tracer, but not throw exceptions), I think it should be fine. Some testing would be needed.