[R6RS] "read/write equivalence"

[R. Kent Dybvig]

> Alan Watson correctly pointed out that it's impossible to require this
> in a sensible way, essentially because of
>
> (eqv? +nan.0 nan+0) => unspecified
>
> I've weakened to spec of `put-datum' accordingly, essentially by
> changing the "must" to a "should".

Why not say "must" and list +nan.0 as an exception?  "should" is too
strong because it slaps the hands of an implementation that has two
representations for NaN, and "should" is too weak because it opens the
door for all kinds of other exceptions.

Kent