fix: memory leak when publishing messages

[plamut]

Fixes #395.

If publish threads are marked as daemonic, the leak seemingly disappears.

How to verify

Set up a topic and a subscription, install memoryprofiler. Add the @profile decorator to the main() function from the sample code and run it (with and without the fix):

$ mprof run sample.py

After each run, plot the memory usage graph:

$ mprof plot

The graphs should look similar to the ones posted here.

PR checklist

• Make sure to open an issue as a bug/issue before writing your code! That way we can discuss the change, evaluate designs, and agree on the general idea
• Ensure the tests and linter pass
• Code coverage does not decrease (if any source code was changed)
• Appropriate docs were updated (if necessary)

[plamut]

Server 500 when running samples, re-starting.

[plamut]

Found a true cause of the issue, it's a bug in CPython.

[plamut]

Fix typo

[pradn]

The fix is good, but I wonder if we'd run into issues by setting these threads as daemons.

The entire Python program exits when no alive non-daemon threads are left.

So, this would mean that scripts that may have created a publisher client, published a bunch of messages, and had only exited when the commit threads (non-daemon) finished running - these scripts could now terminate earlier. However, this may not be a problem, because if users want to guarantee publishes happened, they need to wait for the publish futures to finish with success.

Anyway, just wondering what your thoughts were here.

[pradn]

Peter and I spoke offline: possible premature script termination b/c the threads are no daemons doesn't break a customer guarantee, because as long as the user hasn't received a completed future, they can't assume the publish succeeded.

[pradn]

LGTM

[plamut]

Since it's already late into the workweek, I'll cut a release containing this on Monday.