remove -i and -I, add -m and describe it. Simplify the docs. The user only needs to care about -m, really

I'm not sure if we should go into more detail about how the whole thing works

Fixes #1062

of maintaining magic hex-bit values that must match.

Fixes #1042

Use a mark&mark strategy to test all obj.* bans even if there are
req.* bans in the way.

started fixing that, and then I fixed another couple of nits and polished
the naming a bit and ...

Well, you know how it is.

Changed the backend matching function to take a call-back function
to eliminate the need for the memory which got leaked.

This meant that backend.list could also use it, if only the parsing
code accepted a blank input.

So I rewrote the parsing code to also allow empty names, so you can:
	backend.list (:80)
	backend.list (192.168)
etc.

And then I added a summary report of the probing, if it is happening,
and explicitly say that no probing is happening if it isn't.

Make it possible to mark backends as sick or healthy by way of

  backend.set_health $backendname (sick|healthy|auto)

document gzip'ed.

Fixes	#1029

Fixes	#1046

Also optimise out a few strlen calls in favour of just tracking the
lengths.

Fixes: #1047

Conflicts:

	bin/varnishd/cache_fetch.c

one byte reads for the hex-length headers.

Conflicts:

	bin/varnishd/cache_fetch.c

Conflicts:

	bin/varnishd/cache_esi_fetch.c
	bin/varnishd/cache_fetch.c
	bin/varnishd/cache_gzip.c

catch trouble that materializes only when we destroy the VGZ instance.

Fixes	#1037
Fixes	#1043
Fixes	#1044

Conflicts:

	bin/varnishd/cache.h
	bin/varnishd/cache_center.c
	bin/varnishd/cache_esi_deliver.c
	bin/varnishd/cache_esi_fetch.c
	bin/varnishd/cache_fetch.c
	bin/varnishd/cache_gzip.c
	bin/varnishd/cache_response.c