Currently we use bufio.Reader.ReadLine which accepts either \n or \r\n
as line ending. That is however not correct:

- we should not accept e.g. "S'abc'\r\n." pickle, because it is
  invalid:

	In [32]: pickle.loads(b"S'abc'\r\n.")
	---------------------------------------------------------------------------
	UnpicklingError                           Traceback (most recent call last)
	<ipython-input-32-b1da1988bae1> in <module>()
	----> 1 pickle.loads(b"S'abc'\r\n.")

	UnpicklingError: the STRING opcode argument must be quoted

- we should not accept e.g. "L123L\r\n.", because it is also invalid:

	In [33]: pickle.loads(b"L123L\r\n.")
	---------------------------------------------------------------------------
	ValueError                                Traceback (most recent call last)
	<ipython-input-33-7231ec07f5c4> in <module>()
	----> 1 pickle.loads(b"L123L\r\n.")

	ValueError: invalid literal for int() with base 10: '123L\r\n'

- treating \r as part of EOL in e.g. UNICODE pickle would just drop encoded
  information:

	# python
	In [34]: pickle.loads(b"Vabc\r\n.")
	Out[34]: 'abc\r'

  while ogórek currently decodes it as just 'abc' (no trailing \r).

For this reason let's fix Decoder.readLine to treat only \n as EOL.

Besides this fix, we now get another property: previously, when internally
using bufio.Reader.ReadLine we were not able to distinguish two situations:

- a line was abruptly ended without any EOL characters at all,
- a line was properly ended with EOL character.

Now after we switched to internally using bufio.Reader.ReadSlice, we will be
able to properly detect EOF and return that as error. This property will be
needed in the following patch.