[test] Wait for AsyncTraceWriter to finish sending traces

[marcotc]

This PR addresses an old, but recently more pronounced flaky test, regarding a Datadog::Workers::AsyncTraceWriter test.

This this was trying to wait until the writer had flushed all data, but instead, was waiting until the writer had started processing all traces (by calling #work_pending?).

#work_pending? is used internally by the writer to know if there are traces queued to be process, thus allowing the tracer to either go on a busy processing loop, or sleep for a short bit. This behaviour describes correctly if "Is there more work to be done next?", which is exactly what it represents, but does not answer "Is all the work done?".

This PR waits for the tracer to completely finish its work before verifying the desired conditions.

There's no impact on the correctness of the AsyncTraceWriter, as it is behaving as expected.

[codecov-io]

Codecov Report

Merging #1422 (1a29c0b) into master (c0762e7) will decrease coverage by 0.00%.
The diff coverage is 0.00%.

@@            Coverage Diff             @@
##           master    #1422      +/-   ##
==========================================
- Coverage   98.13%   98.13%   -0.01%     
==========================================
  Files         771      771              
  Lines       36929    36930       +1     
==========================================
  Hits        36242    36242              
- Misses        687      688       +1     

Impacted Files	Coverage Δ
spec/ddtrace/workers/trace_writer_spec.rb	95.18% <0.00%> (-0.28%)	⬇️
lib/ddtrace/patcher.rb	96.66% <0.00%> (-3.34%)	⬇️
...b/action_pack/action_controller/instrumentation.rb	95.71% <0.00%> (+1.42%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c0762e7...1a29c0b. Read the comment docs.

[ivoanjo]

Big thanks for fixing this. I had seen it, but ended up not looking into it, but was really getting annoyed. Much ❤️ .

[ivoanjo]

I'm a big fan of self-describing code. If we like the comment, why not turning this into def more_items_to_be_proccessed? or similar? :)

[marcotc]

I do agree that comments <<< well-named-things.

I mean, isn't "more_items_to_be_processed?" =~ "work_pending?" 😄 I think the main issue is that, like all things complex, there's nuance that would be hard to capture in the name.

Maybe "more_items_to_be_processed?" or similar would be better, but I'm sneaking away with the ✅ for now for the non-flaky CI.