This patch reworks busyobj handling so that busyobjs are owned by the issuing worker, and the worker should explicitly release it when done with it. A busy objcore only lends a pointer to it. This is to fix the current situation where the owning party is murky, and the current code will leak busyobjs in at least one situation.

BusyObj's are reference counted in preperation for the streaming code.

Patch also adds several asserts to help make sure the busyobjs are sound.

Fixes: #1068

Revert two lines that got last in previous commit.

Prompted to by:	DocWilco

backend under very rare circumstances.

distinguish it from the electables (do_esi, do_gzip etc).

steps.

A couple of minor polishishes.

The fetching worker has wrk->busyobj set

Push is_g[un]zip flags into busyobj as proof of concept.