pass, it will be anyway.

Fixes #1335

The only real trouble was purges, and they now have their own vcl_purge{}

Sorry for the detour...

(Don't tell me nothing good ever comes from DB's delayed trains...)

a 304 response, but we still don't get the body.

These are stackable, and does stuff on the way out.

The bottom one points to HTTP/1 ie. WRW for now.

On top of that we can push Range: processing.

Modularizing this, makes it trivial to enable Range: support for
delivery-side gunzip.

(Committed above the Elbe)

for all deliveries, streaming or non-streaming

off stored objects.

We used to need this in struct obj, so we could munge obj.http, but
we don't allow that any more.

that in obj.

Here be dragons.   For instance that if a HEAD is passed, we should
let the Content-Length: unmolested through, whereas normally C-L is a
property of the protocol delivery of the HTTP response.

into req_fsm.c::cnt_deliver{}

have fetched their body.

in HSH_Lookup()