are handled so fast that there may be no numerical difference between
the start and end session timestamps.

This is a consequence of our decision to use C type double (ieee 64bit FP)
for timestamps, at the current distance from POSIX epoch, 1.5 billion
seconds, the resolution is 238 nanoseconds.

Patch from daghf (with minor changes)

Submitted by:	fgs

as priv's inside ESI's delivery code.

delivered.

Transports are a superset of protocols because internal requests,
notably ESI:includes, doesn't use a HTTP protocol to deliver their
response, but have have ødipal ways of doing things.

Submitted by: Dridi

it to slim down req/busyobj a tiny bit.

its cheaper to eliminate the exceedingly unlikely resource leak
Coverity spotted that way.

Fixes #1692

Found and diagnosed by:	martin
Slightly different patch by me.

it is supected of being the cause of the race in #1675 amongst other
evils.

With backend waiters it makes no sense to bring it back.

Fixes: #1726

In future versions of HTTP requests can be terminated abnormally
without ditching the entire session, and this change starts the
transition to that logic.

Fixes: #1732

and plunk it down in req_fsm for now.

Only allow VDP_ungzip to set a length if it is the top VDP.