After it was reported on #varnish that following the previous
example may lead to invalid response headers.

present in the desired workspace.

I'm a little bit uneasy about the tacit assumption that arguments
to VRT_String() now belong to VRT_String(), but we'll see how it
goes.

Fixes: #2132

VRT_CollectString() fail the VCL on overflows.

VRT_StringList() can be used in the rare and probably
non-existent cases where overflow-marking is undesired.

More work on vcl4.1 reserved words etc.

In vcl 4.1 symbol names are unique across all types & kinds.

We want it to be very explicit when non-default VCL versions
are used in test-cases.

For now, the only case this adds is a missing leading zero in the
day of the month of RFC822/RFC1123 - style http dates as in

		Fri, 2 Mar 2018 14:26:02 GMT (bad)
instead of	Fri, 02 Mar 2018 14:26:02 GMT (correct)

This is clearly invalid according to
https://tools.ietf.org/html/rfc7231#section-7.1.1.2
yet accepted by other webserver software.

because the messages in the reference version were horrible.

Also don't rely on protocol numbers, they vary per platform.