Portpicker

* Resolve an internal source of potential flakiness on the bind/close port
checks when used in active environments by calling `.shutdown()` before
`.close()`.

* Add `-h` and `--help` text to the command line tool.
* The command line interface now defaults to associating the returned port
with its parent process PID (usually the calling script) when no argument
was given as that makes more sense.
* When portpicker is used as a command line tool from a script, if a port is
chosen without a portserver it can now be kept bound to a socket by a
child process for a user specified timeout. When successful, this helps
minimize race conditions as subsequent portpicker CLI invocations within
the timeout window cannot choose the same port.
* Some pylint based refactorings to portpicker and portpicker\_test.
* Drop 3.6 from our CI test matrix and metadata. It probably still works
there, but expect our unittests to include 3.7-ism's in the future. We'll
*attempt* to avoid modern constructs in portpicker.py itself but zero
guarantees. Using an old Python? Use an old portpicker.

* Do not re-pick a known used (not-yet-returned) port when running stand alone
without a portserver.

* When not using a portserver *(you really should)*, try the `bind(0)`
approach before hunting for random unused ports. More reliable per
https://github.com/google/python_portpicker/issues/16.

* Add portserver support to Windows using named pipes. To create or connect to
a server, prefix the name of the server with `` (e.g.
`unittest-portserver`).

* Use `async def` instead of `asyncio.coroutine` in order to support 3.10.
* The portserver now checks for and rejects pid values that are out of range.
* Declare a minimum Python version of 3.6 in the package config.
* Rework `portserver_test.py` to launch an actual portserver process instead
of mocks.