Real.maxFinite bug

[Henry Cejtin]

I wouldn't expect it to fail to terminate because of the 80 bit extended format,
but I would expect it to return the wrong answer.  The 80 bit format not only
has a few extra mantissa bits, but also expect it to give the wrong answer.
Why would one use this goofy test in Real_isFinite?  The C library defines
finite (in math.h) to do exactly this and it does it the correct way.  It is
portable, and also compilers must do the right thing for funny architectures
like the x86.