Increase in errors due to no active logging span

[clokep]

Starting on Oct 1 (which corresponds with a re-deploy of matrix.org) we've been seeing a lot more of the following error:

There was no active span when trying to inject the span into a byte dict. Did you forget to start one or did a context slip?

https://sentry.matrix.org/sentry/synapse-matrixorg/issues/15795/

[clokep]

I believe that this is a regression in the range of bbb7ca1f157f43250a2373b194c7dee3dfbfca80 to 7b6f857aa996ec4dde368e11115bf1684770d78f.

[clokep]

Based on the message, it seems this is from inject_active_span_byte_dict, which is used in only two spots:

synapse/synapse/http/matrixfederationclient.py

Line 408 in 1c26243

inject_active_span_byte_dict(headers_dict, request.destination)

synapse/synapse/replication/http/_base.py

Line 205 in 866c84d

inject_active_span_byte_dict(headers, None, check_destination=False)

I've seen this on workers not handling federation, which makes me guess this is from replication?