hybrid text/binary mount

[DJ Delorie]

> I think that most output is done in more than one character at a time
> mode.

Yeah, but consider, for example, raw PBM files.  The first few lines
are text lines, but after that the data is binary.  One could easily
assume that it's a text file from the first few writes, but it's not.
Only if the user uses stdio, and the buffer happens to be big enough
(and not flushed, or line buffered), will it guess right.

I'm mostly trying to cater to the cases where cygwin programs
interact.  In auto mode, a program should accept any type of file
correctly, and produce a file that any cygwin program will interpret
correctly.  It's not an ideal solution (due to interactions between
cygwin programs and non-cygwin programs) but it will cover many of the
common cases.  Programs that really care should use non-auto mode and
interpret the file their own ways, as usual.