Make close synchronous

[puddly]

This PR makes close truly synchronous by:

1. Not calling ser.flush() when closing the serial port. This is a breaking change but I don't think it was correct to wait for flush to be called anyways.
2. Calling ser.close() within the event loop, not in an executor thread. I believe this should be safe given the implementation. All of the file descriptors are opened in non-blocking mode.

For background on point 2, see pyserial#82. I believe that this codepath (TCP socket:// URIs) is now properly special cased and handled within pyserial-asyncio-fast. At this point, pyserial should be used only for serial ports.

Fixes #7.

[cottsay]

What does this mean for SerialTransport.abort()?

Isn't the intended difference between close() and abort() simply that abort() doesn't flush but close() does?

[puddly]

Good point.

The semantics for WriteTransport.abort make it sound like it's harsher than close:

Close the transport immediately, without waiting for pending operations to complete. Buffered data will be lost. No more data will be received. The protocol’s protocol.connection_lost() method will eventually be called with None as its argument.

WriteTransport.close waits for data to be flushed.

It may just be me but it seems like a common assumption for transports to be closed when you call close. It seems like the correct behavior is to wait for the connection_lost event.

[puddly]

I think #16 is the right fix. This approach only works for POSIX and maintains the same bug in event ordering.