exponent spill again

[David Hough 754R work <754r@xxxxxxxxxxx>]

It looks like I should have spelled out the ramifications of my
definitions a little better:

* The extremely broad definition of exponent spill, 
signaling as many cases as possible, is:

 the magnitude of a non-zero finite unrounded result is larger than
 the largest normalized number in the destination format
 or smaller than the smallest normalized number in the destination format.

The signal does not depend on rounding mode.
For overflow, this definition has not been specified by 754 or 754R, 
or implemented - it might be difficult to implement efficiently.
This definition corresponds to 754R's underflow with tininess 
detected before rounding, and to 754's underflow with tininess detected
before rounding and underflow trap enabled.     

* The extremely narrow definition of exponent spill, 
signaling as few cases as possible, is:

 the numerical result rounded to the destination format differs from the
 numerical result rounded to a hypothetical format with the same precision
 as the destination format but unbounded exponent range.

The signal depends on rounding mode.
This definition corresponds to 754R's overflow.
This definition is not part of 754R underflow.
This definition corresponds to 754's underflow
with tininess detected after rounding, loss of accuracy detected
by normalization loss, and underflow trap disabled, 
and has not been implemented - it might be difficult to implement efficiently.