This is a cosmetic tweak, it doesn't change any behavior. No need to
bump the serial for that. The benefits are a better output of silent
make executions, and the preservation of configured variables like
$(PYTHON) all the way downstream.

A word on this stunt:

    \$(A""M_V_GEN)

The backslash is here to escape the dollar sign, nothing too surprising.
However the expression between the brackets is rejected by autoconf
because it looks like an undefined macro. To prevent that, the string is
closed and reopen at once to break the token-inside-brackets construct.

This makes it possible to load VMOD shared objects into a debugger
after a varnishd crash.

fixes https://github.com/varnishcache/varnish-cache/issues/2313

The parsing still requires some TLC but this does not make it any
worse so let's get this in first.

Prompted by and ideas from github::xcir via #2077.

When inserting a new hpack table entry using a referenced name, and
the referenced name would be removed from the table as a result of the
insertion, there was a bug in the memory move/copy operations that
would memmove a too large area. This wouldn't cause problems because
we never used the full 32 bytes that the standard sets as per element
overhead.

Fix it, and the same time simplify the logic to be easier to read. Add
a test case for trimmed indexed name insertion where the table isn't
empty after removing the indexed entry.

Fixes #2307

This is a cherry-pick of 3271c51e that
fixed the issue for Varnish 3.0 since it wasn't a problem in the 4.1
series then. This is only a back-port of the regression test since the
referenced commits already fixed the regression.

Refs 5e4d6c84
Refs 9f422620
Refs #2069

This should help diagnose #2253.

It should have been done for 5.1.2 but hasn't. This is a minor bump of a
single addition so the impact is limited. I also left the release date
to "unreleased" on purpose, since it technically hasn't been released as
6.1 yet.

of days ... just in case.

already failed.

Fixes #2301

Fixes #2247

Fixes #2304