This test routine claimed in a comment to disable flow control. It
actually requires the serial port to be transparent for all 256 byte
values, which in turn requires deactivation of software flow control,
because otherwise the byte-transparency tests fail for bytes 0x11
(Ctrl-Q, XON) and 0x13 (Ctrl-S, XOFF) on systems such as Ubuntu Linux
where software flow control is active by default.

Added new functions to set and clear cumulative timeouts for blocking
read and write functions, a new `Timeout` exception that will be
thrown, and time-tracking code in calls to the blocking read/write
functions. This way, users can now set overall timeouts on other IO
functions, such as `readuntil`, that make multiple calls to our
blocking functions.

That what other close(::IO) methods in Base do. It looks quite odd in the
REPL to get the entire closed SerialPort object thrown back at you.

also polished some docstrings and fixed a parameter name inconsistency

The low-level API wrappers now sit in a submodule LibSerialPort.Lib.

The high-level API currently still re-exports a few symbols from that
low-level API module, to preserve backwards compatibility.

This is due to

* my previously started attempt to merge the low and high-level
  APIs for the three functions sp_flush, sp_drain, and sp_output_waiting

* the fact that several high-level APIs currently still refer to
  enum types and constants defined by the low-level API

We may want to phase out either or both in future, for a cleaner
separation, and a more Julian high-level API.

Arm64 builds have long filed, therefore let's allow them to fail in Travis until that problem is fixed, such that PR authors for other issues do not get build errors that they are not responsible for.