It turns out in core/PyFile.java TextWrapper.read() always returns '\n' if
it finds a '\r', even on platforms where separator is not "\r\n", even when
the character immediately follows is not '\n'. It did unread(), but the
return value was still the replaced '\n'.
I added a boolean sep_is_crlf, renamed and fixed the initial assignment to
Tested the bug case on both Windows and Linux.
Tested Lib/test/test_b2.py relevant portion.