[NNTP] STREAMING diffs (take 2)

[Ade Lovett]

Russ Allbery wrote:
> One question is do we want to say something in the protocol description
> about pre-commit caching, or do we want to leave that to a followup BCP?

I don't see why.  It's entirely possible to write a CHECK/TAKETHIS
module simply using the yes/no responses, as opposed to the "431 maybe".

Indeed, for purely internal article transmission, when machines are
connected via considerably higher-speed pipes than at large, the
overhead of dealing with receiving articles twice or more may be
considerably less than the code complexity of handling the "maybe" at
both receiving and, more likely, sending ends.

Pre-commit caches are an implementation optimization around making
transit boxes deal with insane article-per-second transaction rates, and
as such are best left to a BCP.

Having the distinction between the protocol document and the BCP is
important to my mind, since the latter could change quite significantly
with no change in the former.

-aDe