This reverts commit d771942c, it needs
to be discussed first.

OSX/64 issue...

Related to: #1773

we don't yet know the length.

I attempted to add a way to determine if the object was big enough
*yet* to satisfy the Range, but that was non-viable.

Fixes:	#1777

Don't install cache_backend.h and cache_director.h, they expose more
than what a VMOD needs to see.

The VRT API jumps directly from struct vrt_backend to struct director,
which makes struct backend more than simply opaque, it is completely
invisible. However struct director needs to be visible and has been
moved to cache.h along with its functions definitions.

... like 'Unknown keyword: $Init' for 4.0 vcc files.

Make sure probes can be managed outside of the cli thread, now that
backends can be created any time.

Fixes:	#1775

Fixes: #1776

Add placeholders for pending patches that would need to be documented if
merged.

timestamp syscall in very rare circumstances.

This saves an entire state in the H1 FSM. (At the cost of not
differentiating between read-ahead and full pipelining in VSC.)

Inspired by patches from Lasse & Martin

This makes it easier to run multiple instances, each one writing to
different files.

This commit cleans up a number of weeds and loose ends:

The complete_function should not muck about with the pipelining or WS
reservation.

Only when we have dissected the headers for good do we know how long
the headers were.

Make the HTTP1_Dissect* functions have consistent parameters.