[django] Fix gunicorn gevent worker causing database connection unpatch

[willgittoes-dd]

When running django via gunicorn and using gevent workers new connections are created after a greenlet fork, and they are unpatched. That results in still seeing django traces but losing DB traces.

This simply makes connection patching slightly more robust, by patching the ConnectionHandler object. The patched ConnectionHandler makes sure that every connection it handles is patched.

[willgittoes-dd]

Sorry, un-assigning reviewers for now because I thought about a better way to do this, although testing it is going to suck.

[willgittoes-dd]

Actually my 'better method' was not possible. I wanted to not patch the ConnectionHandler but patch the Database to ensure new connections get the TracedCursor added. But gevent creates a new Database object for each greenlet, so I do indeed need to patch ConnectionHandler as I originally did.

[palazzem]

It should address the problem. Can we replicate the issue with/without this patch in a real application?

[willgittoes-dd]

Yes, I can replicate this with a real (toy) application, not just in tests. That's how I developed the fix. Then I added tests (ensuring the fail without the fix, of course).

Repro steps:

1. Take the Datadog APM django example project: https://github.com/DataDog/trace-examples/tree/master/python/django
2. Run with gunicorn gettingstarted.wsgi:application --bind localhost:8080 --worker-class sync
3. Observe django and defaultdb traces
4. Run with gunicorn gettingstarted.wsgi:application --bind localhost:8080 --worker-class gevent
5. Observe only django traces.