When no -j argument is specified, walk the list to get the most
capable jail technology which can run in the given circumstances.

Due to our somewhat spendthrift ways with argument letters, it
is not realistic to repurpose -[sS]andbox or -[pP]rivsep argument
letters, so the terminology ended up being "jail" because -j was free.

Right now no jails are implemented, and the code does nothing.

Timeout_req was measured from t_idle(which could be either the moment
the connection was accepted or the time of the last activity of the
previous request on the same connection)

Timeout_req, according to the definition(varnishd/mgt/mgt_param_tbl.c)
is now measured starting from t_first(timestamp for the
first byte received).

both introduced yesterday.

already set up in mgt_main.

This state could be set by the expiry timer even though a reference is
grabbed during lookup, causing the assertion to trigger.

Fixes: #1539

Store C-src and compiled VCL under their VCL names, rather than
randomized names -- we know the VCL names are unique.

Spotted by fgs.

Anything that would be put here should instead
be added to Future_* in the wiki.

Anyone upgrading from 1.x to 2.0 today has greater problems than
what this section could help them with.

but fail the fetch.

Fixes #1672

Based mostly on patch from martin

(...or something.  Needless to say, somebody with autocrap clue should
stare at this in disbelief.)

This is very unlikely to make any difference, but the sheer stupidity
fot he PTHREAD_MUTEX_NORMAL definition is ubearable, now that I've
become aware of it.

Inspired by: CERT secure coding POS04-C

received timestamp) instead of sp->t_idle.

Spotted by: gquintard

portability of our backtrace hack.