VCL_STRANDS is an alternative to VCL_STRING_LIST which uses a
struct instead of varargs.

This means that multiple VCL_STRANDS can be passed to a function,
and string-compare really needed that to not waste workspace.

Add <, >, <=, >= string comparisons while here anyway.

The struct and a const char *[N] array are built on the stack and
they are only valid for the duration of the function call.

Add a new type "STRINGS" which is internal to vcc_expr.c only,
and convert to "STRING" and "STRING_LIST" at the edge of vcc_expr.c
and as necessary inside.  (More to come.)

Various and sundry cleanups along the way.

Otherwise some compilers *cough* GCC *cough* may fail to see that it's
properly written before any read.

it is genereally useful also for vmods

STRING_LIST to STRING for error messages.

Fixes #2510

Submitted by:	klean@github

Give procedures proper data structures and stick the VCL methods into
the symbol table up front and GC a lot of small hacks and workarounds.

Timing out an H/2 session needs to take into account any active streams
being processed.

that VCL_ENUM have canonical pointers.

involved versions of same with a simple pointer comparison.

by various standardsbodies.

Going forward these macros will all be named 'v_[a-z_]*_' and only
vdef.h gets to define them.

Renames:
	__match_proto__()		->	v_matchproto_()
	__v_printflike()		->	v_printflike_()
	__state_variable__()		->	v_statevariable_()
	__unused			->	v_unused_
	__attribute__((__noreturn__))	->	v_noreturn_

As a side-effect of the last one, the rule that "vdef.h" be the
first non-<...> include in .c files, possibly via a subsystem
.h (such as vct.h) needs to be enforced harder.

(autocraps "config.h" still needs to go before <...> includes)

Fixes #2508

Fixes #2506

it was deprecated since varnish 5.1.

motivated by the bereq.retries cleanup, see #2405