make != and > and >= fallbacks work for all types

[StefanKarpinski]

This seems simple, but there's some complication that made the obvious change fail.

[StefanKarpinski]

This allows dropping a few definitions in string.j and a few other places.

[JeffBezanson]

!= is there already.

The problem with > is that number types rely on fallbacks for both converting > to < and for promotion, but only one of those can be the most specific. The definition

>(x,y)=(y<x)

is shadowed by

>(x::Real, y::Real) = (>)(promote(x,y)...)

for all Real types, so the promotion definition calls itself in a loop. To fix this we might have to change it to

>(x::Real, y::Real) = (<)(promote(y,x)...)

[StefanKarpinski]

How about we just get rid of the

>(x::Real, y::Real) = (>)(promote(x,y)...)

rule since its redundant given

>(x::Real, y::Real) = (>)(promote(x,y)...)

and the more general

>(x,y) = y < x

? And similarly for >=, of course.