celery: use strongrefs for celery signals

[Kyle-Verhoog]

This hopefully addresses #1011

There's really no reason for us to not use strong references for our signals.

[jd]

I'm really dubious on this one. The fact that celery might use weakref should not change anything since the connect functions are not going to be garbage collected anytime. So their binding should not disappear.

At least if this is the case, we should have some sort of test that proves that this fixes anything.

(I don't have a problem with logging more, maybe do a separate PR?)

[Kyle-Verhoog]

I agree that changing the refs doesn't make sense but it did solve the issue for a user.

I attempted to get a regression test by using import gc; gc.collect() at various points in my Django/Celery app but couldn't get a repro. Even if we did get a repro, I'm not sure if it would be deterministic.

It should be fine to make this change (no need for weak refs anyway), we just don't have support for regressions unfortunately 😕.

I'll make a note to revisit once we get the time.

[shuvozula]

NOTE: Changing the refs did make the Celery traces show up in the UI, when used with ddtrace.contrib.django