have started to fill it.  Add a param to the STV_trim() method
to indicate acceptability of move.

The only reason I don't remove STV_trim() entirely, is that a site
with gazillions very small objects, will want to disable streaming
and use malloc to squeeze as many objects into memory as possible.

with Req/Resp/Bereq/Beresp/Obj instead.

on everybody agreeing about the order of these fields.

oc->busyobj pointer.

Don't deref the objcore until the busyobj dies.

Apply two workarounds for now:  Don't return non-busy object with
a busyobj in HSH_Lookup(), and trade the busyobj ref for an objcore ref
in cnt_fetchbody()

Getting closer...

OC's will only have two states: BUSY or not BUSY.

A BUSY OC may not yet have a busyobj, in which case we assume that
it vary-matches.

Replace the busy_oc pointer with flag, all we are about is the existence
of a busy OC, we're not going to actually touch it.

objcore assignment.

and use of t0.  Always initialize t0 to something.

Correctly spotted by:	FlexeLint

to touch the (potential) new objhead, unless we intend to insert it.

Straighten out the colvolved logic of critbit while here.

The MTX_UNLOCK debug lines will log for how long this thread
held the lock.

The MTX_LOCKWAIT debug lines will log how long the thread waited for the
lock to be acquired.

Bugs: Threads unlocking the mutex as part of pthread_cond_wait will
not update the time of lock value for that mutex on pthread_cond_wait
entry.

stevedore for the default case (malloc and file)

Change the type of oc->priv2 from unsigned to uintptr_t (due to
padding this does not increase the size of objcore).

Change the default_oc_getlru to find the stevedore through the
oc->priv2 instead of having to go through (and page-in) the object.

VRY only needs req, not sess.
Overlooked some lingering sp->wrk's in HSH_Lookup()

Now that we have busyobjs, we do not need to return the objhdr
separately.

Introduce a local req variable to reduce sp->req dereferences.