Ref: #2402

My previous patch did not update the manual and the usage message for
varnishtest concerning the default internal buffer size.

Since SunOS needs a litt bit more workspace_thread, this test case
needs to be updated.

The default buffer size for vtc (configurable with -b) had to be
increased.

Fixes #2402.

Closes #2407

a STRING (const char *) to vmod_priv->priv (void *) (const qualifier).

References #2402

The locking around the use of vcl_active does not include the checking
of its magic value or the temperature asserts, leading to a race when
changing vcl_active.

Fixes: #2390

The struct vcl and the temperature rwlock were not free'd on
discarding labels.

We can directly use vmod-purge now.

This module offers a finer-grained control over purges, in the old
pre-vcl 4.0 style. For that, the signatures of VRT_purge and HSH_Purge
were updated to inform about the number of rearmed objects.

Fixes #2403.

This one is for h2 streams, so the thread pools needed to learn about
"stream" tasks. To properly test this counter, thread_pool_timeout is
now disregarded when the vtc_mode debug flag is set. A minimum value
of 10s is unrealistic for the test suite, it is now hard-coded to 0.5s
when ran through varnishtest.

A client can safely retry the request because no application logic is
involved when no worker can enter the VCL state machine. At this point
the HPACK block was properly processed, so Varnish remains in sync with
the client and may process future streams.

Fixes #2311

This patch fixes the queuing logic of tasks. Depending on the various
thread pool parameters, it could exceed up to `thread_pool_max` tasks
in the queue.

Fixes #2409.

Remove the unmap-handling from vsm_refresh_set.

Revert bdd70290 "Always munmap(2) segments when we delete them."

	#include <sys/mman.h>
Move them back.

Related to #2305

Spotted by:	fgs