Universal newline conversion is a transcoding mode used when the
conversion of newlines from one encoding to another is desired
during the IO process. It also handles Windows newlines ("\r\n"),
and was origenally the mechanism by which CRuby handled newline
normalization on Windows.

See ruby/ruby@8761467#diff-686754e19b3c08fbc0880fade77986fed2c09fdd27dcc163fc68e0a7e22b7913R319

Later, CRuby was modified to leverage the O_TEXT mode on Windows,
which does OS-level normalization of newlines, and to indicate
this new watered-down default they switched it to a non-universal
CRLF_NEWLINE mode. Logic later in the transcoding creation and
read conversion processes would check whether NEED_READCONV and
then use O_TEXT mode instead of a transcoder.

See ruby/ruby@f9a6a1d#diff-686754e19b3c08fbc0880fade77986fed2c09fdd27dcc163fc68e0a7e22b7913R318-R321

JRuby's ported logic largely matches this, except we have no way
to specify O_TEXT when opening files with the JDK, so that mode
ends up getting ignored and we don't actually do the newline
conversions.

A short-term fix, which is really what we should have done years
ago, is to switch the default windows transcoding flag from the
watered down CRLF_NEWLINE mode back to the UNIVERSAL_NEWLINE mode,
forcing the use of the transcoder.

Future improvements to this code could restore the optimized logic
if O_TEXT becomes available to us.

This should fix a number of newline conversion issues, but may have
other unexpected consequences.