No wonder gzip support leaked memory...

Orginally this memory came from the the thread workspace so deallocation
was automatic.  Not so when we switched to malloc...

use them for VSM allocation

Spotted by:	Geoff

(Did you remember that toupper() can be called with EOF ?)

cold of spring and add the "duration" conversion function to vmod_std.

The way these conversions work is that you give them the string you
want converted, and the value to use if they cannot.

like so:

	set beresp.ttl = std.duration(beresp.http.x-ttl, 10 s);

Use system libjemalloc if found, if not we use the bundled one.

uuidgen lives in e2fsprogs which is almost always installed, unlike
mkpasswd which lives in expect, so change to using uuidgen.  No change
for installed systems.

How to deal with Varnish crashing (syslog, etc) and how to find
503 errors in Varnishlog.

Also added Kristians workaround for dealing with errors

http_resp_size is allocated on the thread_pool_workspace.  This means
we will hit assert errors if http_resp_size is larger than the
thread_pool_workspace.  8k should be enough in most cases, from a
light sample of web sites in the wild.

Thanks to askalski.

closes #867

Keep is for future use by conditional backend requests.

req.ttl provides a way to force a fetch of a not yet stale object,
see how in testcase c00042.

We can check the size of the return value too.

I forgot to rename the tv_get variable to tv in configure.ac

Fixes #853

(again)

Fixes: #853

it no longer saves us memory.

rather than saving the entire thing.

This also moves a bit more of the fetch-policy/standards-compliance into
rfc2616.c

accordingly.