Let clients of streaming workunit handler consume independent workunit streams

[gshuflin]

We currently have a global max verbosity for workunits handled by the StreamingWorkunitHandler system. Instead, individual handlers should be able to individually request which level of workunits they want to receive.

[Eric-Arellano]

Related, switch each callback to run as a separate thread. The main benefit is that some callbacks could finish async, and others synchronously, whereas currently all must be the same.

However, this is blocked by:

to poll independently from multiple threads you’d need to change the rust-side polling method to keep multiple streams of workunits, or have one thread that polled and then sent copies to the two handlers or something

each call to “poll_workunits” has a sideeffect of consuming everything

[Eric-Arellano]

See #11692 for an example of how to get the thread-per-callback working on the Python side. However, this suffers from the above problem with Rust's poll_workunits being destructive and not safe with multiple threads.