Fix a couple of issues around calling SSHReader.readuntil() with a timeout

This commit fixes a problem where data could be discarded when calling
SSHReader.readuntil() or SSHReader.readline() wrapped in a call to
asyncio.wait_for() with a timeout. If partial data was received which
didn't include the delimiter but the call then timed out, the partial
data was previously getting discarded since it was already removed from
the buffer. With this change, data remains in the receive buffer until
the delimiter is seen.

This commit also addresses a race condition where an excception could be
raised when issuing a read() immediately after a timeout occurred on a
prior read if the new read is executed before the coroutine which
issued the previous read had a chance to be scheduled. Now, instead of
returning an exception when simultaneous reads are issued, an asyncio
lock is used to serialize multiple simultaneous reads on a stream.
Order of simultaneous reads is not guaranteed, but each read will wait
for others to complete before starting so that the data returned is not
interleaved across them.

An issue still remains that readuntil() may block indefinitely if
the receive window fills up before a delimiter is sent. This will be
addressed in a future commit.