we want to delay destruction until we have delivered all of the
worklist, but we need to fini the vdps in the same order as
varnish-cache serial esi.

the error case is gzgz which needs to update the parent at the right
point in time

using the paren't was wong

in this context, req is the unpender, but we need to push to the crc
node's parent

See comments in the test about unfortunate limits using vtc to
test gzipped H2 responses.

Repeating the test in e00019.vtc with HTTP/2 fails both here and
for standard Varnish. My understanding of HTTP/2 is too poor for
me to know whether there is a bug in a Varnish, or if I am not
applying HTTP/2 correctly. More details in the comments of the vtc.

Setting this aside for now, but we should come back, but we should
come back to it (and if necessary file a Varnish bug report).

because the push phase of unpending happens outside the tree lock, we
need to ensure only one thread is ever entering unpending

this is in preparation for future changes, right now, unpending is only
ever called from the topreq thread

This is the first step towards bringing back safe front pushes: For a
thread in vdp_push bytes, it is safe to push upwards if it owns all
ST_OPEN T_NEXUS nodes upwards or if they are ST_CLOSED or ST_DELIVERED.

more to come in later commits