Refs 1970515d
Refs 0f935cce

Restructured so that:

* 'Upgrading' is limited to work that has to be done to upgrade from
  a current deployment to the new version.

* 'Changes' is a comprehensive, user-level description of changes and
  new features.

Conflicts:
	doc/sphinx/whats-new/index.rst

Ref 30bfe667
Ref #3528

this is a varnishtest test

CID 1643158

CID 1643159

When there is no proxy, server.ip is the same as local.ip

If a shutdown command is last in the vtc client, it closes the file descriptor
right after the shutdown, such that varnishd can not reliably send the GOAWAY,
which in turn changes the H2 error message.

Ref #4283
Ref #4284
Closes #4293

Trying to understand if this is the place where #4293 happens

Closes #4283

We no longer send "real" H2CE error codes when the connection is simply EOF or
timed out, but we did still emit SessError.

We now limit it to unhandled HTC status and add to the additional Debug output
the HTC status.

It is fine for a client to close the connection, we should say a friendly
goodbye and log this as no error.

Closes #4281
Related to #4283

A newly created request's t_first is now when the first (partial) read of the
HEADERS frame is complete.

t_req is now when all headers are complete.

This should match http1 more closely

Implements #4282

someone(tm) added some unmotivated logic way back in
1e82f502:

As long as the VDP list is not empty, there is a first VDP on the list, so why
would we have statements like AN(vdpe) or check if (vdpe != NULL) ?

Diff best viewed ignoring whitespace (git log -pb)

The existing code suggested that HEADERS could arrive on an existing stream,
which was then checked for its state to be idle.

But because we do not support PUSH_PROMISE, the only way to open a new stream is
by receiving a HEADERS frame, which opens a new stream (see rfc9113 section
5.1). In our case, the "idle" state for all streams is implicit.

Diff is best viewed ignoring whitespace (git log -pb)

Just moving code, no change

Now guess which change is going to be proposed...

Just moving code, no change

otherwise, for the modified vtc, a continuation is still expected when another
stream is opened:

1031 SessError       c H2: expected continuation but  received HEADERS on stream 3
1031 Debug           c H2 CLEANUP H2CE_PROTOCOL_ERROR

Partly addresses #4283

Connection and stream errors are not the same thing, so our error reporting
should also use unambiguous names.