Reduce notifications for a minor speedup

[x42005e1f]

Avoids calling slow notification methods when there are no waiters.

[x42005e1f]

In fact, false positive cases (the number of waiters is higher than it actually is) are possible, since the thread must enter the finally block to decrement. But access to the waiting queue requires access to the undocumented Condition attribute (_waiters, both async and sync), which may break in future versions of Python or other implementations of the standard library, so it is not used in this PR. Also, due to the delayed effect of notifications (as they are performed via new tasks and threads), it is basically impossible to obtain actual information unless complex cancellation handling is added. Therefore, the number of notifications is not optimal, but still safe.

[codecov]

Codecov Report

Attention: Patch coverage is 96.72131% with 4 lines in your changes missing coverage. Please review.

Project coverage is 94.42%. Comparing base (ad2b1bd) to head (7b404db).
Report is 3 commits behind head on master.

Files with missing lines	Patch %	Lines
tests/test_mixed.py	93.65%	2 Missing and 2 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #704      +/-   ##
==========================================
- Coverage   94.44%   94.42%   -0.02%     
==========================================
  Files           5        5              
  Lines        1369     1472     +103     
  Branches      109      127      +18     
==========================================
+ Hits         1293     1390      +97     
- Misses         52       55       +3     
- Partials       24       27       +3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[codspeed-hq]

CodSpeed Performance Report

Merging #704 will improve performances by 78.94%

_{Comparing x42005e1f:master (7b404db) with master (ad2b1bd)}

Summary

⚡ 4 improvements

Benchmarks breakdown

	Benchmark	master	x42005e1f:master	Change
⚡	test_bench_async_put_sync_get	1,089.9 ms	125.9 ms	×8.7
⚡	test_bench_sync_put_async_get	1,150.2 ms	636.9 ms	+80.59%
⚡	test_bench_sync_put_async_join	1,196.7 ms	255.8 ms	×4.7
⚡	test_sync_join_async_done	1,167.2 ms	652.3 ms	+78.94%

[asvetlov]

Thanks, the speedup looks very impressive!
I'll take a careful review after sleeping on the PR.

[asvetlov]

@x42005e1f thank you for adding more tests! I'm out of my laptop right now, I'll review the PR again when I come back.

[asvetlov]

Looks good, thanks!