Attempt to fix a flaky test on Windows

[njsmith]

This failure:
https://ci.appveyor.com/project/njsmith/trio/build/1.0.768/job/3lbdyxl63q3h9s21
appears to be due to some floating point wackiness causing a >=
assertion to not quite succeed. See comment for more details.

[codecov]

Codecov Report

Merging #361 into master will decrease coverage by 2.59%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #361      +/-   ##
==========================================
- Coverage   99.24%   96.64%    -2.6%     
==========================================
  Files          87      174      +87     
  Lines       10437    20874   +10437     
  Branches      728     1456     +728     
==========================================
+ Hits        10358    20174    +9816     
- Misses         61      623     +562     
- Partials       18       77      +59

Impacted Files	Coverage Δ
trio/tests/test_timeouts.py	100% <100%> (ø)	⬆️
trio/_core/_windows_cffi.py	0% <0%> (-91.31%)	⬇️
trio/_core/tests/test_windows.py	23.33% <0%> (-76.67%)	⬇️
trio/_core/_io_windows.py	0% <0%> (-75.74%)	⬇️
trio/_util.py	87% <0%> (-6%)	⬇️
trio/_highlevel_open_tcp_listeners.py	94.73% <0%> (-5.27%)	⬇️
trio/_socket.py	96.1% <0%> (-3.9%)	⬇️
trio/_core/__init__.py	97.14% <0%> (-2.86%)	⬇️
trio/tests/test_socket.py	98.24% <0%> (-1.76%)	⬇️
trio/_core/tests/test_io.py	98.87% <0%> (-1.13%)	⬇️
... and 91 more

---

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 a5d0417...65a59f9. Read the comment docs.

[njsmith]

Codecov complaints here are due to a bug in codecov failing to combine the windows and non-windows runs. I emailed them about it and they say they've added a workaround. We'll see, but in the mean time I think this is OK to go in :-)