Patch by:	scoof
Fixes		#961

as per discussions in httpbis WG

different functions.

I have not managed to write a vtc case for this one, but I am pretty
sure this:

Fixes:	#945

a21746d2.

Fixes: #941

of source code.

This is substantially DocWilcos fix, but instead of adding yet another
unique numbering variable, I have collapsed the once we have into
a single one.

Fixes #948

Use it to hold the derived vary string for the request (if any) and
skip busy objects with non-matching Vary during hash lookup.

Idea by:	sky

Not detected by:  FlexeLint 9.00f
Detected by:  FlexeLint 9.00g

to prevent old junk on the workspace from being taken for valid.

cache_hash to cache_center.

do not trusting) that all objects under an objhdr have the same
Vary string.

Build a vary string for the request on the wrk->ws as we do the
comparison and use it as a cache for subsequent comparisons.

This will also be first step of sky's "pre-vary"

white space.

Repack struct object, reducing a couple of field widths along the way.

This shaves 16 bytes of struct object.

We're seeing at least up to 4k in the wild, so increase the limit
somewhat.

See: #939

Print out the rest of the format, changing the error from

  "Unknown format character: {"

to:

  "Unknown format starting at: %{asdf}x"

Fall back to "-" if the handling has not been decided yet.

Fixes: #950
Fixes: #944
Fixes: #918

utils expect to receive SIGHUP).

Fixes: #947

memlevel 8.

Fix testcases which depend on older values.

gzip_window and gzip_memlevel.

Still same pessimal values though.

Fixes: #876