value: use stable sort in binary iota

Otherwise the index values will be unpredictable if there are duplicates.

I believe this

Fixes robpike#118

ivy: allow for placeholder indexes, standing for the dimension at tha…

…t point

If we want to extract a block inside a matrix, say the 2nd column
of a 3x3 matrix, we can write

	x[iota 3; 2]

but one must know the dimension ahead of time. When programming,
one can use rho to pull this out, but it's messy.

This change provides a shorthand: an empty index expression evaluates
to iota "dimension here". Thus the above can be shortened to

	x[;2]

Empty components can be in any location, so

	x[;;]

is the same as x in our example. (So is x[;], but that's a little
more subtle.) And because of how ivy works already

	x[2;]

is the same as

	x[2]

On the drive by, allow an origin to be any non-negative value.

ivy: fix matrix index of matrix (robpike#112)

The right-to-left evaluation order for indexes was causing the
accumulation of the shape to end up reversed. Fix that by doing
an additional reverse after the loop.

Fixes robpike#111.

value: fix bug in big integer exponentiation

Rationals use this routine, which could overwrite the exponent.

Fixes robpike#108

ivy: add the arc-hyperbolics asinh etc.

demo: add complex numbers to the demo

Very briefly, but they deserve a mention.

value: make Int constants actual constants

No reason not to, and it takes a few bytes out of the binary.

ivy: make sure internal uses of operators cannot be overridden by use…

…r-defined ones

There were only a few actually used through EvalUnary and EvalBinary, and it was
easy to avoid using them, and probably more efficient anyway.

Fixes robpike#107

value: s/pi/π/ in a comment

ivy: add log and power for complex values

Also make phase a method on Complex, like abs, and do a little
renaming for consistency.

Also fix a trivial bug in complex cosine: delete a "Neg" dreg.