feat: Close anomalous spans in Datadog middleware

[timmc-edx]

Also, add an additional assertion in run_middleware to ensure that the middleware hasn't silently entered an error state. (Helped me debug a case where I had forgotten to fake a Waffle flag.)

Merge checklist:
Check off if complete or not applicable:

• Version bumped
• Changelog record added
• Documentation updated (not only docstrings)
• Fixup commits are squashed away
• Unit tests added/updated
• Manual testing instructions provided
• Noted any: Concerns, dependencies, migration issues, deadlines, tickets

[timmc-edx]

I wasn't able to do a completely faithful manual test because we don't have a way of generating the anomalous spans. However, I was able to do something sort of close by mangling the code a bit:

1. Get the middleware installed and the feature activated in devstack
2. Run LMS without DD_DJANGO_INSTRUMENT_MIDDLEWARE=false to ensure there's a fat stack of spans to work with, and DD_TRACE_DEBUG=true to confirm that spans really are being closed
3. Modify close_anomalous_spans() to skip the duration check, use while walk_span.resource != 'edx_django_utils.monitoring.DeploymentMonitoringMiddleware.__call__': rather than looking for a span with name django.request, and print the current span at the end.

With that, it appears that the middleware above the arbitrarily selected one are finished but the selected one and the ones below are not (until the trace finishes naturally).