Test chunked

Reduce number of repeats because of vtc_log limit

I suspect that gcov might not register properly when the process terminates with
a panic?

Ref #4366

There was a hole in it since we made vxid_t 64bit

(gdb) ptype /o struct vsl_log
/* offset      |    size */  type = struct vsl_log {
/*      0      |       8 */    uint32_t *wlb;
/*      8      |       8 */    uint32_t *wlp;
/*     16      |       8 */    uint32_t *wle;
/*     24      |       4 */    unsigned int wlr;
/* XXX  4-byte hole      */
/*     32      |       8 */    vxid_t wid;

                               /* total size (bytes):   40 */
                             }

Likely every storage engine needs it

When the loop completes, the loop variable is now set to NULL. This allows to
easily differentiate between a completion and a break, avoids an out-of-bounds
value after the loop and matches the vqueue macros.

d-oh, I undid a fix which I applied and explained myself before in
1157dfdc

This reverts commit d9d38175.

Refs #3352

Ref #3352

This reverts commit eab881fc.

Adding the path apparently was wrong, I now suspect that the actual cause was
the one addressed by bf75bb69

Since 362c24c3 we did not flexelint the teken
code, and accidentally brought it back with
48e21c3c because in vtest{,2} it no longer lives
in a subdirectory.

Fixes #4357

If even if we do not hand hdl->last to the caller to return because we are in a
finished or error bos state, we are still done with it and need to clear it for
re-use.

Fixes #4356

The current usage of obj_boc_notify_Unlock() was always following a call
to Lck_Lock(), and never in a context where the BOC mutex was already
held.

Conflating the broadcast and unlock operations prevents conditional
notifications from the #4073 candidate.

since the merge, the great diversity of our vtest machines exposed a vai lease
leak via c00111.vtc:

In storage_simple.c, we can not free the last struct storage in the storage list
before we potentially advanced to the next. The previous code was failing to
return it.

We now use a null_iov pointer with a zero length viov to pass the lease to the
iterator with the only purpose to return it.

This is now the first time that we use lib sources from vtest2 which are not in
varnish-cache.

Background: vudp has been moved into its own source files from vtc_syslog.c

I personally think we should allow it in general for the error handling pattern,
but if anything it should be ok in vtest2

Truncate the uint64_t interpreted as a pointer first, then cast to an actual
pointer.

So yes, 32bit still keeps us real...

I had accidentally committed some WIP which should have been on the stash

This reverts parts of commit 1c7db798.

Ref 37040b4d

No idea when this popped up, but I suppose it must have come with an update of
the system flexelint runs on

File /usr/include/sys/queue.h, Line 425
            _
	    "not head's first field address",				\
    Warning 607: Parameter 'head' of macro found within string

(and others)

cache/cache_gzip.c:264:17: error: pointer targets in assignment differ in signedness [-Werror=pointer-sign]
cache/cache_gzip.c:266:18: error: pointer targets in assignment differ in signedness [-Werror=pointer-sign]
cache/cache_gzip.c:282:16: error: pointer targets in assignment differ in signedness [-Werror=pointer-sign]