Fix reloader on OSX py38 and Windows

[Tronic]

Simplified a bit and this now appears to work on all OS. On Windows we don't get graceful shutdowns though, due to lack of SIGTERM.

Also removes app.run(..., **kwargs) that was silently consuming invalid arguments, and forces using keywords for all args but host and port (potentially breaking some apps).

[codecov]

Codecov Report

Merging #1827 into master will increase coverage by 2.62%.
The diff coverage is 85.00%.

@@            Coverage Diff             @@
##           master    #1827      +/-   ##
==========================================
+ Coverage   92.17%   94.79%   +2.62%     
==========================================
  Files          23       24       +1     
  Lines        2314     2307       -7     
  Branches      424      421       -3     
==========================================
+ Hits         2133     2187      +54     
+ Misses        141       75      -66     
- Partials       40       45       +5     

Impacted Files	Coverage Δ
sanic/reloader_helpers.py	75.40% <82.35%> (+60.67%)	⬆️
sanic/app.py	93.02% <100.00%> (+0.76%)	⬆️
sanic/__main__.py	71.87% <0.00%> (ø)

---

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 4658e0f...b636083. Read the comment docs.

[Tronic]

This includes #1555 but due to bigger and potentially breaking changes will need more testing. One known issue is that if the server does not exit, the autoreloader will get stuck waiting until another Ctrl+C, after which it will probably abandon the server process and any worker processes that it may have, because this no longer uses external tools to kill all children. This should be fixed in server code, making it exit and terminate its children in a more reliable manner on SIGTERM (for instance, making non-graceful exit if not completed within a timeout). SIGKILL on the server is not a good idea, especially so in multiple worker mode, where the workers would be left running.

I believe that eventually autoreloader should be merged to the main process of multiworker mode but that will require a lot more work. Another consideration is that with workers=1 possibly the main process should run the server and the autoreloader be its child, to allow normal debugging etc.

[ahopkins]

I have tested the PR to make sure it still works on Linux. I have not pulled it to confirm in OSX or Windows yet.