Small Arrow Spec update

[Brian Rice]

I will be busy for a few days, but here's a quick note to get your minds
working on the arrow idea.  I mentioned that there wasn't a clear framework
for getting at arrow meta-information (i.e. arrows referring to a given
arrow), but it turns out to be quite simple: simply invert the arrows of
car-graph and/or cdr-graph, and apply this inverse to an arrow to obtain
all arrows referring to that one, with elementary classification accoring
to whether it's the car or cdr reference.

Also, there is the extension of 'apply' to take entire graphs and have
their arrows collectively churned by another graph acting as a function.
However, there are many ways to do this, so the default semantics I will
not state until I am sure of it.  Note also that graphs are state-machines,
and that as such the 'apply' mechanism encapsulates all updates to the
machine-state from the given one.

So, what remains is a good picture of the system that the basic arrow
framework can reflect on, and use this as our graph-construction framework.

	-Brian