with references.

The last three commits already made configure recommend installing
Python 3 packages and look for versioned executables, however with a low
priority.

This is a problem on systems such as OpenBSD 6.5 with a default Python
version at 2.7, where 3.7 flavored Python packages get installed with
a "-3" binary suffix.  That is, when both rst2man and rst2man-3 are
installed at configure time, the lower version will be picked unless
explicitly passed through `--with-feature' arguments.

Regardless of this specific case, trying more specificly versioned tool
names first seems correctly in line with recent development and less
error prone, so change it accordingly.

This reverts commit 86af5ce0.

Unfortunately we have two scopes top and task that may conflict when
esi_level is zero. This restores the id field, making struct vrt_priv
slightly larger than when it used to be managed as a tailq. But in
the context of a red-black tree for dynamic priv lookups the benefits
should still outweigh the slight increase in memory footprint.

Conflicts:
	bin/varnishd/cache/cache_vrt_priv.c

Fixes #3003

A regression from d7a81fe8 and the
followup changes from #2993. The new VSS_Resolve{One,First} functions
expect a non-null address string.

Done with git tricks involving:

    sed '
        s:NEEDLESS(return NULL):NEEDLESS(return (NULL)):
	s:NEEDLESS(return(:NEEDLESS(return (:
    '

Unfortunately it won't work with a Coccinelle patch, so I couldn't
update return.cocci to handle the NEEDLESS case.

This breaks the API since some functions didn't take a VRT_CTX until now.

Closes #2860

returns version related strings.

Get rid of global, but pretty useles variable VCS_version.

Closes: #2936

And sync the documentation with the current behavior, part of which used
to be implicit.

This is how std.ip is documented, so VSS_ResolveOne doesn't work there.
It might not be the only migration to VSS_ResolveOne that requires
attention.

Speaking of attention, VSA_Malloc may require some.

It can now be shared by established callback-based resolvers and the new
VSS_ResolveOne. This also changes the semantics of VSS_ResolveOne in the
sense that the port is now a default port, overriden by the address if
it contains one.

Also make it clear that VTCP was already relying on a VSS function that
didn't and still doesn't allow a null errp argument, while conversely
all VTCP_open call sites provide a valid errp argument.

At this point the workspace already overflowed so there's no point
making further progress to fail on the next workspace operation.

Incidentally, we are failing for an IP conversion, not an integer.

We do not hold a reference, the magic can be unstable.

The name implies that this is not for production usage.

There are a surprising number of subtle corner cases here, and I
think I have handled them all correctly, but caveat emptor...

With recent changes in the VCL state and temperature handling this test
case lost its relevance and started testing a no-op operation. And since
it wasn't apparently obvious what it was trying to check between two
possible interpretations it now tests both and better verifies it.

Reported by fgs.

When I initially wrote this test case it used to trigger a panic in the
then young VCL labeling.