It seems that browser writers also have trouble figuring out that
twisted logic in HTTP/1.1.

Fixes:	#1404
Spotted by:	Scoof

actually wanted to do.

Hat-tip:	scn

Fixes	#1395

for the bg-fetch, before we release the request to carry on.

Fixes	#1399

the last fixe was clearly not enough.

Fixes	#1398

(And Happy New Year!)

bother with compat-crap for srandomdev()

srandomdev() any more

Hit by:	scoof

headers, we need to purge all traces of it, before we construct our
503 response, otherwise we risk failing to set important headers,
such as "Connection: close".

Fixes #416
(again)

rather than just the ttl.

Fixes		#1268
Spotted by:	daghf
Partial fix by:	gquintard

is not the same as a positive TTL any more.

"shortlived" test puts the object in Transient.

Patch by:	gquintard
Fixes		#1387

Fixes #1323
Spotted & fix by:	gquintard

we cannot just throw the storage away, we have to wait for the
fetch-thread to go away, possibly in response to a new "abandon"
signal.

Spotted first by:	scoof
Fixes	#1391

right order:  Init is done first to last "inifin", Fini is done
in opposite order, last to first.

This replaces the several ad-hoc facilities we had.