20241028
  On machines with arithmetic conforming to IEEE standard 754 (most
modern machines)

	option allow_NaN 2;

now permits handling Infinity and NaN (Not a Number) as specified in
the IEEE standard, whereas

	option allow_NaN 1;

turns Infinity into NaN and the default

	option allow_NaN 0;

causes computations that would involve NaNs and Infinities to give
error messages.  "option allow_NaN 1" has long been available;
"option allow_NaN 2" is new.  For example, if file krf defines and
uses a rational function

	# See W. Kahan (1987), "Presubstitution, and Continued Fractions"
	# http://www.arithmazium.org/classroom/lib/Kahan_Presubstitution_and_Continued_Fractions.pdf
	var x;
	var f = 4 - 3/(x - 2 - 1/(x - 7 + 10/(x - 2 - 2/(x-3))));
	for{i in 1..4} {
		let x := i;
		display x, f;
		}

then by default "ampl krf" says

	x = 1
	Error at _cmdno 3 executing "display" command
	(file f/krf, line 7, offset 255):
	error computing defined variable f:
		can't compute 10/0

but the input "option allow_NaN 2; include krf" gives

	x = 1
	f = 7

	x = 2
	f = 4

	x = 3
	f = 1.6

	x = 4
	f = 2.5

and "option allow_NaN 1; include krf" gives

	x = 1
	f = NaN

	x = 2
	f = NaN

	x = 3
	f = NaN

	x = 4
	f = NaN

If file zork contains

	option allow_NaN 2;
	display exp(3000);
	option allow_NaN 1;
	display exp(3000);
	option allow_NaN 0;
	display exp(3000);

then "ampl zork" gives

	exp(3000) = Infinity

	exp(3000) = NaN

	Error at _cmdno 6 executing "display" command
	(file zork, line 6, offset 98):
		can't evaluate exp(3000): Numerical result out of range

  Fix some bugs with a defined variable involving a piecewise-linear
term applied to another defined variable that involves another
piecewise-linear term, possibly indirectly.  Examples:

	var x1 >= -100 := -100;
	var x2 >= -100 := -100;
	var x3 = 3*x1 - 2*x2 + 100;
	var x4; x4def: x4 = 0.5*(3*x1 - 2*x2 - 100 + <<0;-1,1>>x3);
	var x5 = 3*x1 + 2*x2 - x4 + 9;
	minimize o: 0.5*(3*x1 + 2*x2 + x4 + <<0;-1,1>>x5);
	option solver cplex; solve;
	CPLEX 20.1.0.0: Constraint _scon[1] is not convex quadratic since it is an equality constraint.

With the bug fixed, "solve;" says

	CPLEX 20.1.0.0: optimal integer solution; objective -104.5

Another, similar but more troublesome example is the above with x4
as a defined variable, i.e.,

	var x1 >= -100 := -100;
	var x2 >= -100 := -100;
	var x3 = 3*x1 - 2*x2 + 100;
	var x4 = 0.5*(3*x1 - 2*x2 - 100 + <<0;-1,1>>x3);
	var x5 = 3*x1 + 2*x2 - x4 + 9;
	minimize o: 0.5*(3*x1 + 2*x2 + x4 + <<0;-1,1>>x5);

After the fixes, "option solver cplex; solve;" says

	CPLEX 20.1.0.0: optimal integer solution; objective -104.5
	2 MIP simplex iterations
	0 branch-and-bound nodes

  Fix a bug (corrupted del_mblk arg) with ord(i) when i is a dummy
variable.  The bug appeared in a complicated example; some simple
examples work correctly.

  Allow "from", "to" and some more obscure values to appear as
dummy variables.  Example:

	set A dimen 2; var x{(to,from) in A: to != from};

previously elicited a syntax error message.  Now it is accepted.